diff --git a/docker/docker-compose.yml b/docker/docker-compose.yml
index 2aa6608e..adaecc53 100644
--- a/docker/docker-compose.yml
+++ b/docker/docker-compose.yml
@@ -4,6 +4,23 @@ services:
     build:
       context: .
       args:
+        # Requirements file to use: 
+        # | GPU | CPU | requirements file to use |
+        # |--------|---------|---------|
+        # | NVIDIA | has AVX2 | `requirements.txt` |
+        # | NVIDIA | no AVX2 | `requirements_noavx2.txt` |
+        # | AMD | has AVX2 | `requirements_amd.txt` |
+        # | AMD | no AVX2 | `requirements_amd_noavx2.txt` |
+        # | CPU only | has AVX2 | `requirements_cpu_only.txt` |
+        # | CPU only | no AVX2 | `requirements_cpu_only_noavx2.txt` |
+        # | Apple | Intel | `requirements_apple_intel.txt` |
+        # | Apple | Apple Silicon | `requirements_apple_silicon.txt` |
+        # Default: requirements.txt`
+        # BUILD_REQUIREMENTS: requirements.txt
+        
+        # Extension requirements to build: 
+        # BUILD_EXTENSIONS: 
+        
         # specify which cuda version your card supports: https://developer.nvidia.com/cuda-gpus
         TORCH_CUDA_ARCH_LIST: ${TORCH_CUDA_ARCH_LIST:-7.5} 
         BUILD_EXTENSIONS: ${BUILD_EXTENSIONS:-}
diff --git a/docker/nvidia/Dockerfile b/docker/nvidia/Dockerfile
index 8da55643..6429d5e7 100644
--- a/docker/nvidia/Dockerfile
+++ b/docker/nvidia/Dockerfile
@@ -3,6 +3,7 @@ FROM nvidia/cuda:12.1.1-devel-ubuntu22.04 as builder
 WORKDIR /builder
 ARG TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST:-3.5;5.0;6.0;6.1;7.0;7.5;8.0;8.6+PTX}"
 ARG BUILD_EXTENSIONS="${BUILD_EXTENSIONS:-}"
+ARG BUILD_REQUIREMENTS="${BUILD_REQUIREMENTS:-requirements.txt}"
 ARG APP_UID="${APP_UID:-6972}"
 ARG APP_GID="${APP_GID:-6972}"
 # create / update build env
@@ -18,7 +19,7 @@ RUN --mount=type=cache,target=/root/.cache/pip,rw \
 USER app:app_grp
 # build wheels for runtime
 WORKDIR /home/app/build
-COPY --chown=app:app_grp requirements.txt /home/app/build
+COPY --chown=app:app_grp "$BUILD_REQUIREMENTS" /home/app/build/requirements.txt
 COPY --chown=app:app_grp extensions /home/app/build/extensions
 RUN --mount=type=cache,target=/root/.cache/pip,rw \
     # build all requirements files as wheel dists
diff --git a/modules/chat.py b/modules/chat.py
index 8ddb7531..dccd20b7 100644
--- a/modules/chat.py
+++ b/modules/chat.py
@@ -510,8 +510,7 @@ def load_latest_history(state):
     histories = find_all_histories(state)
 
     if len(histories) > 0:
-        unique_id = Path(histories[0]).stem
-        history = load_history(unique_id, state['character_menu'], state['mode'])
+        history = load_history(histories[0], state['character_menu'], state['mode'])
     else:
         history = start_new_chat(state)
 
diff --git a/modules/training.py b/modules/training.py
index b0e02400..2830ba07 100644
--- a/modules/training.py
+++ b/modules/training.py
@@ -341,7 +341,7 @@ def do_train(lora_name: str, always_override: bool, q_proj_en: bool, v_proj_en:
 
     # Populate target_modules list with chosen X_proj modules. Llama-based models only atm, non-llama will revert to default behavior.
     def list_target_modules(model_id):
-        if model_id != "llama":
+        if model_id != "llama" and model_id != "mistral":
             return model_to_lora_modules[model_id]
 
         available_modules = {
@@ -517,7 +517,8 @@ def do_train(lora_name: str, always_override: bool, q_proj_en: bool, v_proj_en:
     # == Start prepping the model itself ==
     if not hasattr(shared.model, 'lm_head') or hasattr(shared.model.lm_head, 'weight'):
         logger.info("Getting model ready")
-        prepare_model_for_kbit_training(shared.model)
+        if 'quantization_config' in shared.model.config.to_dict():
+            prepare_model_for_kbit_training(shared.model)
 
     # base model is now frozen and should not be reused for any other LoRA training than this one
     shared.model_dirty_from_training = True
@@ -615,7 +616,8 @@ def do_train(lora_name: str, always_override: bool, q_proj_en: bool, v_proj_en:
             warmup_steps=math.ceil(warmup_steps / gradient_accumulation_steps),
             num_train_epochs=epochs,
             learning_rate=actual_lr,
-            fp16=False if shared.args.cpu else True,
+            fp16=False if shared.args.cpu or shared.args.bf16 else True,
+            bf16=shared.args.bf16,
             optim=optimizer,
             logging_steps=2 if stop_at_loss > 0 else 5,
             evaluation_strategy="steps" if eval_data is not None else "no",
@@ -627,7 +629,7 @@ def do_train(lora_name: str, always_override: bool, q_proj_en: bool, v_proj_en:
             # TODO: Enable multi-device support
             ddp_find_unused_parameters=None,
             no_cuda=shared.args.cpu,
-            use_ipex=True if is_torch_xpu_available and not shared.args.cpu else False
+            use_ipex=True if is_torch_xpu_available() and not shared.args.cpu else False
         ),
         data_collator=transformers.DataCollatorForLanguageModeling(shared.tokenizer, mlm=False),
         callbacks=list([Callbacks()])
diff --git a/presets/Debug-deterministic.yaml b/presets/Debug-deterministic.yaml
index cbe7064f..45d5953f 100644
--- a/presets/Debug-deterministic.yaml
+++ b/presets/Debug-deterministic.yaml
@@ -1 +1,2 @@
 do_sample: false
+top_k: 1
diff --git a/requirements.txt b/requirements.txt
index f5cde1c1..0c4a9452 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -28,34 +28,34 @@ bitsandbytes==0.41.1; platform_system != "Windows"
 https://github.com/jllllll/bitsandbytes-windows-webui/releases/download/wheels/bitsandbytes-0.41.1-py3-none-win_amd64.whl; platform_system == "Windows"
 
 # llama-cpp-python (CPU only, AVX2)
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.25+cpuavx2-cp311-cp311-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.11"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.25+cpuavx2-cp310-cp310-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.10"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.25+cpuavx2-cp39-cp39-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.9"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.25+cpuavx2-cp38-cp38-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.8"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.25+cpuavx2-cp311-cp311-win_amd64.whl; platform_system == "Windows" and python_version == "3.11"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.25+cpuavx2-cp310-cp310-win_amd64.whl; platform_system == "Windows" and python_version == "3.10"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.25+cpuavx2-cp39-cp39-win_amd64.whl; platform_system == "Windows" and python_version == "3.9"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.25+cpuavx2-cp38-cp38-win_amd64.whl; platform_system == "Windows" and python_version == "3.8"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.26+cpuavx2-cp311-cp311-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.11"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.26+cpuavx2-cp310-cp310-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.10"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.26+cpuavx2-cp39-cp39-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.9"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.26+cpuavx2-cp38-cp38-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.8"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.26+cpuavx2-cp311-cp311-win_amd64.whl; platform_system == "Windows" and python_version == "3.11"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.26+cpuavx2-cp310-cp310-win_amd64.whl; platform_system == "Windows" and python_version == "3.10"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.26+cpuavx2-cp39-cp39-win_amd64.whl; platform_system == "Windows" and python_version == "3.9"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.26+cpuavx2-cp38-cp38-win_amd64.whl; platform_system == "Windows" and python_version == "3.8"
 
 # llama-cpp-python (CUDA, no tensor cores)
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda-0.2.25+cu121-cp311-cp311-win_amd64.whl; platform_system == "Windows" and python_version == "3.11"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda-0.2.25+cu121-cp310-cp310-win_amd64.whl; platform_system == "Windows" and python_version == "3.10"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda-0.2.25+cu121-cp39-cp39-win_amd64.whl; platform_system == "Windows" and python_version == "3.9"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda-0.2.25+cu121-cp38-cp38-win_amd64.whl; platform_system == "Windows" and python_version == "3.8"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda-0.2.25+cu121-cp311-cp311-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.11"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda-0.2.25+cu121-cp310-cp310-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.10"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda-0.2.25+cu121-cp39-cp39-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.9"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda-0.2.25+cu121-cp38-cp38-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.8"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda-0.2.26+cu121-cp311-cp311-win_amd64.whl; platform_system == "Windows" and python_version == "3.11"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda-0.2.26+cu121-cp310-cp310-win_amd64.whl; platform_system == "Windows" and python_version == "3.10"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda-0.2.26+cu121-cp39-cp39-win_amd64.whl; platform_system == "Windows" and python_version == "3.9"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda-0.2.26+cu121-cp38-cp38-win_amd64.whl; platform_system == "Windows" and python_version == "3.8"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda-0.2.26+cu121-cp311-cp311-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.11"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda-0.2.26+cu121-cp310-cp310-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.10"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda-0.2.26+cu121-cp39-cp39-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.9"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda-0.2.26+cu121-cp38-cp38-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.8"
 
 # llama-cpp-python (CUDA, tensor cores)
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda_tensorcores-0.2.25+cu121-cp311-cp311-win_amd64.whl; platform_system == "Windows" and python_version == "3.11"
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda_tensorcores-0.2.25+cu121-cp310-cp310-win_amd64.whl; platform_system == "Windows" and python_version == "3.10"
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda_tensorcores-0.2.25+cu121-cp39-cp39-win_amd64.whl; platform_system == "Windows" and python_version == "3.9"
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda_tensorcores-0.2.25+cu121-cp38-cp38-win_amd64.whl; platform_system == "Windows" and python_version == "3.8"
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda_tensorcores-0.2.25+cu121-cp311-cp311-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.11"
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda_tensorcores-0.2.25+cu121-cp310-cp310-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.10"
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda_tensorcores-0.2.25+cu121-cp39-cp39-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.9"
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda_tensorcores-0.2.25+cu121-cp38-cp38-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.8"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda_tensorcores-0.2.26+cu121-cp311-cp311-win_amd64.whl; platform_system == "Windows" and python_version == "3.11"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda_tensorcores-0.2.26+cu121-cp310-cp310-win_amd64.whl; platform_system == "Windows" and python_version == "3.10"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda_tensorcores-0.2.26+cu121-cp39-cp39-win_amd64.whl; platform_system == "Windows" and python_version == "3.9"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda_tensorcores-0.2.26+cu121-cp38-cp38-win_amd64.whl; platform_system == "Windows" and python_version == "3.8"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda_tensorcores-0.2.26+cu121-cp311-cp311-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.11"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda_tensorcores-0.2.26+cu121-cp310-cp310-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.10"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda_tensorcores-0.2.26+cu121-cp39-cp39-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.9"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda_tensorcores-0.2.26+cu121-cp38-cp38-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.8"
 
 # CUDA wheels
 https://github.com/jllllll/AutoGPTQ/releases/download/v0.6.0/auto_gptq-0.6.0+cu121-cp311-cp311-win_amd64.whl; platform_system == "Windows" and python_version == "3.11"
diff --git a/requirements_amd.txt b/requirements_amd.txt
index a3d9f90e..7ba9c910 100644
--- a/requirements_amd.txt
+++ b/requirements_amd.txt
@@ -28,20 +28,20 @@ bitsandbytes==0.38.1; platform_system != "Windows"
 https://github.com/jllllll/bitsandbytes-windows-webui/releases/download/wheels/bitsandbytes-0.38.1-py3-none-win_amd64.whl; platform_system == "Windows"
 
 # llama-cpp-python (CPU only, AVX2)
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.25+cpuavx2-cp311-cp311-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.11"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.25+cpuavx2-cp310-cp310-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.10"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.25+cpuavx2-cp39-cp39-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.9"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.25+cpuavx2-cp38-cp38-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.8"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.25+cpuavx2-cp311-cp311-win_amd64.whl; platform_system == "Windows" and python_version == "3.11"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.25+cpuavx2-cp310-cp310-win_amd64.whl; platform_system == "Windows" and python_version == "3.10"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.25+cpuavx2-cp39-cp39-win_amd64.whl; platform_system == "Windows" and python_version == "3.9"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.25+cpuavx2-cp38-cp38-win_amd64.whl; platform_system == "Windows" and python_version == "3.8"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.26+cpuavx2-cp311-cp311-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.11"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.26+cpuavx2-cp310-cp310-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.10"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.26+cpuavx2-cp39-cp39-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.9"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.26+cpuavx2-cp38-cp38-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.8"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.26+cpuavx2-cp311-cp311-win_amd64.whl; platform_system == "Windows" and python_version == "3.11"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.26+cpuavx2-cp310-cp310-win_amd64.whl; platform_system == "Windows" and python_version == "3.10"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.26+cpuavx2-cp39-cp39-win_amd64.whl; platform_system == "Windows" and python_version == "3.9"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.26+cpuavx2-cp38-cp38-win_amd64.whl; platform_system == "Windows" and python_version == "3.8"
 
 # AMD wheels
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/rocm/llama_cpp_python_cuda-0.2.25+rocm5.6.1-cp311-cp311-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.11"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/rocm/llama_cpp_python_cuda-0.2.25+rocm5.6.1-cp310-cp310-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.10"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/rocm/llama_cpp_python_cuda-0.2.25+rocm5.6.1-cp39-cp39-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.9"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/rocm/llama_cpp_python_cuda-0.2.25+rocm5.6.1-cp38-cp38-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.8"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/rocm/llama_cpp_python_cuda-0.2.26+rocm5.6.1-cp311-cp311-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.11"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/rocm/llama_cpp_python_cuda-0.2.26+rocm5.6.1-cp310-cp310-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.10"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/rocm/llama_cpp_python_cuda-0.2.26+rocm5.6.1-cp39-cp39-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.9"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/rocm/llama_cpp_python_cuda-0.2.26+rocm5.6.1-cp38-cp38-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.8"
 https://github.com/jllllll/AutoGPTQ/releases/download/v0.6.0/auto_gptq-0.6.0+rocm5.6-cp311-cp311-linux_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.11"
 https://github.com/jllllll/AutoGPTQ/releases/download/v0.6.0/auto_gptq-0.6.0+rocm5.6-cp310-cp310-linux_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.10"
 https://github.com/jllllll/AutoGPTQ/releases/download/v0.6.0/auto_gptq-0.6.0+rocm5.6-cp39-cp39-linux_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.9"
diff --git a/requirements_amd_noavx2.txt b/requirements_amd_noavx2.txt
index 10e41032..d522910e 100644
--- a/requirements_amd_noavx2.txt
+++ b/requirements_amd_noavx2.txt
@@ -28,14 +28,14 @@ bitsandbytes==0.38.1; platform_system != "Windows"
 https://github.com/jllllll/bitsandbytes-windows-webui/releases/download/wheels/bitsandbytes-0.38.1-py3-none-win_amd64.whl; platform_system == "Windows"
 
 # llama-cpp-python (CPU only, no AVX2)
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.25+cpuavx-cp311-cp311-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.11"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.25+cpuavx-cp310-cp310-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.10"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.25+cpuavx-cp39-cp39-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.9"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.25+cpuavx-cp38-cp38-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.8"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.25+cpuavx-cp311-cp311-win_amd64.whl; platform_system == "Windows" and python_version == "3.11"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.25+cpuavx-cp310-cp310-win_amd64.whl; platform_system == "Windows" and python_version == "3.10"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.25+cpuavx-cp39-cp39-win_amd64.whl; platform_system == "Windows" and python_version == "3.9"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.25+cpuavx-cp38-cp38-win_amd64.whl; platform_system == "Windows" and python_version == "3.8"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.26+cpuavx-cp311-cp311-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.11"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.26+cpuavx-cp310-cp310-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.10"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.26+cpuavx-cp39-cp39-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.9"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.26+cpuavx-cp38-cp38-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.8"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.26+cpuavx-cp311-cp311-win_amd64.whl; platform_system == "Windows" and python_version == "3.11"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.26+cpuavx-cp310-cp310-win_amd64.whl; platform_system == "Windows" and python_version == "3.10"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.26+cpuavx-cp39-cp39-win_amd64.whl; platform_system == "Windows" and python_version == "3.9"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.26+cpuavx-cp38-cp38-win_amd64.whl; platform_system == "Windows" and python_version == "3.8"
 
 # AMD wheels
 https://github.com/jllllll/AutoGPTQ/releases/download/v0.6.0/auto_gptq-0.6.0+rocm5.6-cp311-cp311-linux_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.11"
diff --git a/requirements_apple_intel.txt b/requirements_apple_intel.txt
index 721dc7c0..8132cc4f 100644
--- a/requirements_apple_intel.txt
+++ b/requirements_apple_intel.txt
@@ -28,15 +28,15 @@ bitsandbytes==0.41.1; platform_system != "Windows"
 https://github.com/jllllll/bitsandbytes-windows-webui/releases/download/wheels/bitsandbytes-0.41.1-py3-none-win_amd64.whl; platform_system == "Windows"
 
 # Mac wheels
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.25-cp311-cp311-macosx_11_0_x86_64.whl; platform_system == "Darwin" and platform_release >= "20.0.0" and platform_release < "21.0.0" and python_version == "3.11"
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.25-cp310-cp310-macosx_11_0_x86_64.whl; platform_system == "Darwin" and platform_release >= "20.0.0" and platform_release < "21.0.0" and python_version == "3.10"
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.25-cp39-cp39-macosx_11_0_x86_64.whl; platform_system == "Darwin" and platform_release >= "20.0.0" and platform_release < "21.0.0" and python_version == "3.9"
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.25-cp38-cp38-macosx_11_0_x86_64.whl; platform_system == "Darwin" and platform_release >= "20.0.0" and platform_release < "21.0.0" and python_version == "3.8"
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.25-cp311-cp311-macosx_12_0_x86_64.whl; platform_system == "Darwin" and platform_release >= "21.0.0" and platform_release < "22.0.0" and python_version == "3.11"
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.25-cp310-cp310-macosx_12_0_x86_64.whl; platform_system == "Darwin" and platform_release >= "21.0.0" and platform_release < "22.0.0" and python_version == "3.10"
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.25-cp39-cp39-macosx_12_0_x86_64.whl; platform_system == "Darwin" and platform_release >= "21.0.0" and platform_release < "22.0.0" and python_version == "3.9"
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.25-cp38-cp38-macosx_12_0_x86_64.whl; platform_system == "Darwin" and platform_release >= "21.0.0" and platform_release < "22.0.0" and python_version == "3.8"
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.25-cp311-cp311-macosx_14_0_x86_64.whl; platform_system == "Darwin" and platform_release >= "23.0.0" and platform_release < "24.0.0" and python_version == "3.11"
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.25-cp310-cp310-macosx_14_0_x86_64.whl; platform_system == "Darwin" and platform_release >= "23.0.0" and platform_release < "24.0.0" and python_version == "3.10"
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.25-cp39-cp39-macosx_14_0_x86_64.whl; platform_system == "Darwin" and platform_release >= "23.0.0" and platform_release < "24.0.0" and python_version == "3.9"
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.25-cp38-cp38-macosx_14_0_x86_64.whl; platform_system == "Darwin" and platform_release >= "23.0.0" and platform_release < "24.0.0" and python_version == "3.8"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.26-cp311-cp311-macosx_11_0_x86_64.whl; platform_system == "Darwin" and platform_release >= "20.0.0" and platform_release < "21.0.0" and python_version == "3.11"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.26-cp310-cp310-macosx_11_0_x86_64.whl; platform_system == "Darwin" and platform_release >= "20.0.0" and platform_release < "21.0.0" and python_version == "3.10"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.26-cp39-cp39-macosx_11_0_x86_64.whl; platform_system == "Darwin" and platform_release >= "20.0.0" and platform_release < "21.0.0" and python_version == "3.9"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.26-cp38-cp38-macosx_11_0_x86_64.whl; platform_system == "Darwin" and platform_release >= "20.0.0" and platform_release < "21.0.0" and python_version == "3.8"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.26-cp311-cp311-macosx_12_0_x86_64.whl; platform_system == "Darwin" and platform_release >= "21.0.0" and platform_release < "22.0.0" and python_version == "3.11"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.26-cp310-cp310-macosx_12_0_x86_64.whl; platform_system == "Darwin" and platform_release >= "21.0.0" and platform_release < "22.0.0" and python_version == "3.10"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.26-cp39-cp39-macosx_12_0_x86_64.whl; platform_system == "Darwin" and platform_release >= "21.0.0" and platform_release < "22.0.0" and python_version == "3.9"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.26-cp38-cp38-macosx_12_0_x86_64.whl; platform_system == "Darwin" and platform_release >= "21.0.0" and platform_release < "22.0.0" and python_version == "3.8"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.26-cp311-cp311-macosx_14_0_x86_64.whl; platform_system == "Darwin" and platform_release >= "23.0.0" and platform_release < "24.0.0" and python_version == "3.11"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.26-cp310-cp310-macosx_14_0_x86_64.whl; platform_system == "Darwin" and platform_release >= "23.0.0" and platform_release < "24.0.0" and python_version == "3.10"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.26-cp39-cp39-macosx_14_0_x86_64.whl; platform_system == "Darwin" and platform_release >= "23.0.0" and platform_release < "24.0.0" and python_version == "3.9"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.26-cp38-cp38-macosx_14_0_x86_64.whl; platform_system == "Darwin" and platform_release >= "23.0.0" and platform_release < "24.0.0" and python_version == "3.8"
diff --git a/requirements_apple_silicon.txt b/requirements_apple_silicon.txt
index 42fa7d7c..f813ef22 100644
--- a/requirements_apple_silicon.txt
+++ b/requirements_apple_silicon.txt
@@ -28,19 +28,19 @@ bitsandbytes==0.41.1; platform_system != "Windows"
 https://github.com/jllllll/bitsandbytes-windows-webui/releases/download/wheels/bitsandbytes-0.41.1-py3-none-win_amd64.whl; platform_system == "Windows"
 
 # Mac wheels
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.25-cp311-cp311-macosx_11_0_arm64.whl; platform_system == "Darwin" and platform_release >= "20.0.0" and platform_release < "21.0.0" and python_version == "3.11"
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.25-cp310-cp310-macosx_11_0_arm64.whl; platform_system == "Darwin" and platform_release >= "20.0.0" and platform_release < "21.0.0" and python_version == "3.10"
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.25-cp39-cp39-macosx_11_0_arm64.whl; platform_system == "Darwin" and platform_release >= "20.0.0" and platform_release < "21.0.0" and python_version == "3.9"
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.25-cp38-cp38-macosx_11_0_arm64.whl; platform_system == "Darwin" and platform_release >= "20.0.0" and platform_release < "21.0.0" and python_version == "3.8"
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.25-cp311-cp311-macosx_12_0_arm64.whl; platform_system == "Darwin" and platform_release >= "21.0.0" and platform_release < "22.0.0" and python_version == "3.11"
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.25-cp310-cp310-macosx_12_0_arm64.whl; platform_system == "Darwin" and platform_release >= "21.0.0" and platform_release < "22.0.0" and python_version == "3.10"
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.25-cp39-cp39-macosx_12_0_arm64.whl; platform_system == "Darwin" and platform_release >= "21.0.0" and platform_release < "22.0.0" and python_version == "3.9"
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.25-cp38-cp38-macosx_12_0_arm64.whl; platform_system == "Darwin" and platform_release >= "21.0.0" and platform_release < "22.0.0" and python_version == "3.8"
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.25-cp311-cp311-macosx_13_0_arm64.whl; platform_system == "Darwin" and platform_release >= "22.0.0" and platform_release < "23.0.0" and python_version == "3.11"
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.25-cp310-cp310-macosx_13_0_arm64.whl; platform_system == "Darwin" and platform_release >= "22.0.0" and platform_release < "23.0.0" and python_version == "3.10"
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.25-cp39-cp39-macosx_13_0_arm64.whl; platform_system == "Darwin" and platform_release >= "22.0.0" and platform_release < "23.0.0" and python_version == "3.9"
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.25-cp38-cp38-macosx_13_0_arm64.whl; platform_system == "Darwin" and platform_release >= "22.0.0" and platform_release < "23.0.0" and python_version == "3.8"
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.25-cp311-cp311-macosx_14_0_arm64.whl; platform_system == "Darwin" and platform_release >= "23.0.0" and platform_release < "24.0.0" and python_version == "3.11"
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.25-cp310-cp310-macosx_14_0_arm64.whl; platform_system == "Darwin" and platform_release >= "23.0.0" and platform_release < "24.0.0" and python_version == "3.10"
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.25-cp39-cp39-macosx_14_0_arm64.whl; platform_system == "Darwin" and platform_release >= "23.0.0" and platform_release < "24.0.0" and python_version == "3.9"
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.25-cp38-cp38-macosx_14_0_arm64.whl; platform_system == "Darwin" and platform_release >= "23.0.0" and platform_release < "24.0.0" and python_version == "3.8"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.26-cp311-cp311-macosx_11_0_arm64.whl; platform_system == "Darwin" and platform_release >= "20.0.0" and platform_release < "21.0.0" and python_version == "3.11"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.26-cp310-cp310-macosx_11_0_arm64.whl; platform_system == "Darwin" and platform_release >= "20.0.0" and platform_release < "21.0.0" and python_version == "3.10"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.26-cp39-cp39-macosx_11_0_arm64.whl; platform_system == "Darwin" and platform_release >= "20.0.0" and platform_release < "21.0.0" and python_version == "3.9"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.26-cp38-cp38-macosx_11_0_arm64.whl; platform_system == "Darwin" and platform_release >= "20.0.0" and platform_release < "21.0.0" and python_version == "3.8"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.26-cp311-cp311-macosx_12_0_arm64.whl; platform_system == "Darwin" and platform_release >= "21.0.0" and platform_release < "22.0.0" and python_version == "3.11"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.26-cp310-cp310-macosx_12_0_arm64.whl; platform_system == "Darwin" and platform_release >= "21.0.0" and platform_release < "22.0.0" and python_version == "3.10"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.26-cp39-cp39-macosx_12_0_arm64.whl; platform_system == "Darwin" and platform_release >= "21.0.0" and platform_release < "22.0.0" and python_version == "3.9"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.26-cp38-cp38-macosx_12_0_arm64.whl; platform_system == "Darwin" and platform_release >= "21.0.0" and platform_release < "22.0.0" and python_version == "3.8"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.26-cp311-cp311-macosx_13_0_arm64.whl; platform_system == "Darwin" and platform_release >= "22.0.0" and platform_release < "23.0.0" and python_version == "3.11"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.26-cp310-cp310-macosx_13_0_arm64.whl; platform_system == "Darwin" and platform_release >= "22.0.0" and platform_release < "23.0.0" and python_version == "3.10"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.26-cp39-cp39-macosx_13_0_arm64.whl; platform_system == "Darwin" and platform_release >= "22.0.0" and platform_release < "23.0.0" and python_version == "3.9"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.26-cp38-cp38-macosx_13_0_arm64.whl; platform_system == "Darwin" and platform_release >= "22.0.0" and platform_release < "23.0.0" and python_version == "3.8"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.26-cp311-cp311-macosx_14_0_arm64.whl; platform_system == "Darwin" and platform_release >= "23.0.0" and platform_release < "24.0.0" and python_version == "3.11"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.26-cp310-cp310-macosx_14_0_arm64.whl; platform_system == "Darwin" and platform_release >= "23.0.0" and platform_release < "24.0.0" and python_version == "3.10"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.26-cp39-cp39-macosx_14_0_arm64.whl; platform_system == "Darwin" and platform_release >= "23.0.0" and platform_release < "24.0.0" and python_version == "3.9"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/metal/llama_cpp_python-0.2.26-cp38-cp38-macosx_14_0_arm64.whl; platform_system == "Darwin" and platform_release >= "23.0.0" and platform_release < "24.0.0" and python_version == "3.8"
diff --git a/requirements_cpu_only.txt b/requirements_cpu_only.txt
index 3c815bf9..af9fe3ab 100644
--- a/requirements_cpu_only.txt
+++ b/requirements_cpu_only.txt
@@ -28,11 +28,11 @@ bitsandbytes==0.41.1; platform_system != "Windows"
 https://github.com/jllllll/bitsandbytes-windows-webui/releases/download/wheels/bitsandbytes-0.41.1-py3-none-win_amd64.whl; platform_system == "Windows"
 
 # llama-cpp-python (CPU only, AVX2)
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.25+cpuavx2-cp311-cp311-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.11"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.25+cpuavx2-cp310-cp310-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.10"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.25+cpuavx2-cp39-cp39-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.9"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.25+cpuavx2-cp38-cp38-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.8"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.25+cpuavx2-cp311-cp311-win_amd64.whl; platform_system == "Windows" and python_version == "3.11"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.25+cpuavx2-cp310-cp310-win_amd64.whl; platform_system == "Windows" and python_version == "3.10"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.25+cpuavx2-cp39-cp39-win_amd64.whl; platform_system == "Windows" and python_version == "3.9"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.25+cpuavx2-cp38-cp38-win_amd64.whl; platform_system == "Windows" and python_version == "3.8"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.26+cpuavx2-cp311-cp311-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.11"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.26+cpuavx2-cp310-cp310-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.10"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.26+cpuavx2-cp39-cp39-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.9"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.26+cpuavx2-cp38-cp38-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.8"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.26+cpuavx2-cp311-cp311-win_amd64.whl; platform_system == "Windows" and python_version == "3.11"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.26+cpuavx2-cp310-cp310-win_amd64.whl; platform_system == "Windows" and python_version == "3.10"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.26+cpuavx2-cp39-cp39-win_amd64.whl; platform_system == "Windows" and python_version == "3.9"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.26+cpuavx2-cp38-cp38-win_amd64.whl; platform_system == "Windows" and python_version == "3.8"
diff --git a/requirements_cpu_only_noavx2.txt b/requirements_cpu_only_noavx2.txt
index e9a4b8f1..329963e3 100644
--- a/requirements_cpu_only_noavx2.txt
+++ b/requirements_cpu_only_noavx2.txt
@@ -28,11 +28,11 @@ bitsandbytes==0.41.1; platform_system != "Windows"
 https://github.com/jllllll/bitsandbytes-windows-webui/releases/download/wheels/bitsandbytes-0.41.1-py3-none-win_amd64.whl; platform_system == "Windows"
 
 # llama-cpp-python (CPU only, no AVX2)
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.25+cpuavx-cp311-cp311-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.11"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.25+cpuavx-cp310-cp310-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.10"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.25+cpuavx-cp39-cp39-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.9"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.25+cpuavx-cp38-cp38-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.8"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.25+cpuavx-cp311-cp311-win_amd64.whl; platform_system == "Windows" and python_version == "3.11"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.25+cpuavx-cp310-cp310-win_amd64.whl; platform_system == "Windows" and python_version == "3.10"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.25+cpuavx-cp39-cp39-win_amd64.whl; platform_system == "Windows" and python_version == "3.9"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.25+cpuavx-cp38-cp38-win_amd64.whl; platform_system == "Windows" and python_version == "3.8"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.26+cpuavx-cp311-cp311-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.11"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.26+cpuavx-cp310-cp310-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.10"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.26+cpuavx-cp39-cp39-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.9"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.26+cpuavx-cp38-cp38-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.8"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.26+cpuavx-cp311-cp311-win_amd64.whl; platform_system == "Windows" and python_version == "3.11"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.26+cpuavx-cp310-cp310-win_amd64.whl; platform_system == "Windows" and python_version == "3.10"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.26+cpuavx-cp39-cp39-win_amd64.whl; platform_system == "Windows" and python_version == "3.9"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.26+cpuavx-cp38-cp38-win_amd64.whl; platform_system == "Windows" and python_version == "3.8"
diff --git a/requirements_noavx2.txt b/requirements_noavx2.txt
index a9c9b35f..1608114a 100644
--- a/requirements_noavx2.txt
+++ b/requirements_noavx2.txt
@@ -28,34 +28,34 @@ bitsandbytes==0.41.1; platform_system != "Windows"
 https://github.com/jllllll/bitsandbytes-windows-webui/releases/download/wheels/bitsandbytes-0.41.1-py3-none-win_amd64.whl; platform_system == "Windows"
 
 # llama-cpp-python (CPU only, no AVX2)
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.25+cpuavx-cp311-cp311-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.11"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.25+cpuavx-cp310-cp310-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.10"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.25+cpuavx-cp39-cp39-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.9"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.25+cpuavx-cp38-cp38-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.8"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.25+cpuavx-cp311-cp311-win_amd64.whl; platform_system == "Windows" and python_version == "3.11"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.25+cpuavx-cp310-cp310-win_amd64.whl; platform_system == "Windows" and python_version == "3.10"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.25+cpuavx-cp39-cp39-win_amd64.whl; platform_system == "Windows" and python_version == "3.9"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.25+cpuavx-cp38-cp38-win_amd64.whl; platform_system == "Windows" and python_version == "3.8"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.26+cpuavx-cp311-cp311-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.11"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.26+cpuavx-cp310-cp310-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.10"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.26+cpuavx-cp39-cp39-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.9"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.26+cpuavx-cp38-cp38-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.8"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.26+cpuavx-cp311-cp311-win_amd64.whl; platform_system == "Windows" and python_version == "3.11"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.26+cpuavx-cp310-cp310-win_amd64.whl; platform_system == "Windows" and python_version == "3.10"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.26+cpuavx-cp39-cp39-win_amd64.whl; platform_system == "Windows" and python_version == "3.9"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.26+cpuavx-cp38-cp38-win_amd64.whl; platform_system == "Windows" and python_version == "3.8"
 
 # llama-cpp-python (CUDA, no tensor cores)
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda-0.2.25+cu121avx-cp311-cp311-win_amd64.whl; platform_system == "Windows" and python_version == "3.11"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda-0.2.25+cu121avx-cp310-cp310-win_amd64.whl; platform_system == "Windows" and python_version == "3.10"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda-0.2.25+cu121avx-cp39-cp39-win_amd64.whl; platform_system == "Windows" and python_version == "3.9"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda-0.2.25+cu121avx-cp38-cp38-win_amd64.whl; platform_system == "Windows" and python_version == "3.8"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda-0.2.25+cu121avx-cp311-cp311-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.11"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda-0.2.25+cu121avx-cp310-cp310-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.10"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda-0.2.25+cu121avx-cp39-cp39-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.9"
-https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda-0.2.25+cu121avx-cp38-cp38-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.8"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda-0.2.26+cu121avx-cp311-cp311-win_amd64.whl; platform_system == "Windows" and python_version == "3.11"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda-0.2.26+cu121avx-cp310-cp310-win_amd64.whl; platform_system == "Windows" and python_version == "3.10"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda-0.2.26+cu121avx-cp39-cp39-win_amd64.whl; platform_system == "Windows" and python_version == "3.9"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda-0.2.26+cu121avx-cp38-cp38-win_amd64.whl; platform_system == "Windows" and python_version == "3.8"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda-0.2.26+cu121avx-cp311-cp311-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.11"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda-0.2.26+cu121avx-cp310-cp310-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.10"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda-0.2.26+cu121avx-cp39-cp39-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.9"
+https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda-0.2.26+cu121avx-cp38-cp38-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.8"
 
 # llama-cpp-python (CUDA, tensor cores)
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda_tensorcores-0.2.25+cu121avx-cp311-cp311-win_amd64.whl; platform_system == "Windows" and python_version == "3.11"
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda_tensorcores-0.2.25+cu121avx-cp310-cp310-win_amd64.whl; platform_system == "Windows" and python_version == "3.10"
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda_tensorcores-0.2.25+cu121avx-cp39-cp39-win_amd64.whl; platform_system == "Windows" and python_version == "3.9"
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda_tensorcores-0.2.25+cu121avx-cp38-cp38-win_amd64.whl; platform_system == "Windows" and python_version == "3.8"
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda_tensorcores-0.2.25+cu121avx-cp311-cp311-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.11"
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda_tensorcores-0.2.25+cu121avx-cp310-cp310-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.10"
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda_tensorcores-0.2.25+cu121avx-cp39-cp39-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.9"
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda_tensorcores-0.2.25+cu121avx-cp38-cp38-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.8"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda_tensorcores-0.2.26+cu121avx-cp311-cp311-win_amd64.whl; platform_system == "Windows" and python_version == "3.11"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda_tensorcores-0.2.26+cu121avx-cp310-cp310-win_amd64.whl; platform_system == "Windows" and python_version == "3.10"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda_tensorcores-0.2.26+cu121avx-cp39-cp39-win_amd64.whl; platform_system == "Windows" and python_version == "3.9"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda_tensorcores-0.2.26+cu121avx-cp38-cp38-win_amd64.whl; platform_system == "Windows" and python_version == "3.8"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda_tensorcores-0.2.26+cu121avx-cp311-cp311-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.11"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda_tensorcores-0.2.26+cu121avx-cp310-cp310-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.10"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda_tensorcores-0.2.26+cu121avx-cp39-cp39-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.9"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda_tensorcores-0.2.26+cu121avx-cp38-cp38-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.8"
 
 # CUDA wheels
 https://github.com/jllllll/AutoGPTQ/releases/download/v0.6.0/auto_gptq-0.6.0+cu121-cp311-cp311-win_amd64.whl; platform_system == "Windows" and python_version == "3.11"