Merge dev branch

README.md

@@ -263,6 +263,7 @@ List of command-line flags
 | `--tensor_split TENSOR_SPLIT`       | Split the model across multiple GPUs. Comma-separated list of proportions. Example: 18,17. |
 | `--numa`      | Activate NUMA task allocation for llama.cpp. |
 | `--logits_all`| Needs to be set for perplexity evaluation to work. Otherwise, ignore it, as it makes prompt processing slower. |
+| `--no_offload_kqv` | Do not offload the K, Q, V to the GPU. This saves VRAM but reduces the performance. |
 | `--cache-capacity CACHE_CAPACITY`   | Maximum cache capacity (llama-cpp-python). Examples: 2000MiB, 2GiB. When provided without units, bytes will be assumed. |
 
 #### ExLlama

modules/llamacpp_hf.py

@@ -204,6 +204,7 @@ def from_pretrained(cls, pretrained_model_name_or_path: Optional[Union[str, os.P
             'tensor_split': tensor_split_list,
             'rope_freq_scale': 1.0 / shared.args.compress_pos_emb,
             'logits_all': shared.args.logits_all,
+            'offload_kqv': not shared.args.no_offload_kqv
         }
 
         Llama = llama_cpp_lib().Llama

modules/llamacpp_model.py

@@ -86,6 +86,7 @@ def from_pretrained(self, path):
             'rope_freq_base': RoPE.get_rope_freq_base(shared.args.alpha_value, shared.args.rope_freq_base),
             'tensor_split': tensor_split_list,
             'rope_freq_scale': 1.0 / shared.args.compress_pos_emb,
+            'offload_kqv': not shared.args.no_offload_kqv
         }
 
         result.model = Llama(**params)

modules/loaders.py

@@ -43,6 +43,7 @@
         'compress_pos_emb',
         'cpu',
         'numa',
+        'no_offload_kqv'
     ],
     'llamacpp_HF': [
         'n_ctx',
@@ -63,6 +64,7 @@
         'trust_remote_code',
         'no_use_fast',
         'logits_all',
+        'no_offload_kqv',
         'llamacpp_HF_info',
     ],
     'ExLlamav2_HF': [

modules/shared.py

@@ -117,6 +117,7 @@
 parser.add_argument('--tensor_split', type=str, default=None, help='Split the model across multiple GPUs. Comma-separated list of proportions. Example: 18,17.')
 parser.add_argument('--numa', action='store_true', help='Activate NUMA task allocation for llama.cpp.')
 parser.add_argument('--logits_all', action='store_true', help='Needs to be set for perplexity evaluation to work. Otherwise, ignore it, as it makes prompt processing slower.')
+parser.add_argument('--no_offload_kqv', action='store_true', help='Do not offload the  K, Q, V to the GPU. This saves VRAM but reduces the performance.')
 parser.add_argument('--cache-capacity', type=str, help='Maximum cache capacity (llama-cpp-python). Examples: 2000MiB, 2GiB. When provided without units, bytes will be assumed.')
 
 # ExLlama

modules/text_generation.py

@@ -265,7 +265,7 @@ def apply_stopping_strings(reply, all_stop_strings):
 
 def get_reply_from_output_ids(output_ids, state, starting_from=0):
     reply = decode(output_ids[starting_from:], state['skip_special_tokens'])
-    if hasattr(shared.tokenizer, 'convert_ids_to_tokens') and len(output_ids) > starting_from and shared.tokenizer.convert_ids_to_tokens(int(output_ids[starting_from])).startswith('▁'):
+    if (hasattr(shared.tokenizer, 'convert_ids_to_tokens') and len(output_ids) > starting_from and shared.tokenizer.convert_ids_to_tokens(int(output_ids[starting_from])).startswith('▁')) and not reply.startswith(' '):
         reply = ' ' + reply
 
     return reply

modules/ui.py

@@ -91,6 +91,7 @@ def list_model_elements():
         'rope_freq_base',
         'numa',
         'logits_all',
+        'no_offload_kqv',
         'hqq_backend',
     ]
     if is_torch_xpu_available():

modules/ui_model_menu.py

@@ -115,6 +115,7 @@ def create_ui():
                             shared.gradio['mlock'] = gr.Checkbox(label="mlock", value=shared.args.mlock)
                             shared.gradio['numa'] = gr.Checkbox(label="numa", value=shared.args.numa, info='NUMA support can help on some systems with non-uniform memory access.')
                             shared.gradio['cpu'] = gr.Checkbox(label="cpu", value=shared.args.cpu)
+                            shared.gradio['no_offload_kqv'] = gr.Checkbox(label="no_offload_kqv", value=shared.args.no_offload_kqv, info='Do not offload the  K, Q, V to the GPU. This saves VRAM but reduces the performance.')
                             shared.gradio['load_in_8bit'] = gr.Checkbox(label="load-in-8bit", value=shared.args.load_in_8bit)
                             shared.gradio['bf16'] = gr.Checkbox(label="bf16", value=shared.args.bf16)
                             shared.gradio['auto_devices'] = gr.Checkbox(label="auto-devices", value=shared.args.auto_devices)

requirements.txt

@@ -26,14 +26,14 @@ 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-0.2.23+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-0.2.23+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-0.2.23+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-0.2.23+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-0.2.23+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-0.2.23+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-0.2.23+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-0.2.23+cpuavx2-cp38-cp38-win_amd64.whl; platform_system == "Windows" and python_version == "3.8"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.24+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/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.24+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/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.24+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/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.24+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/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.24+cpuavx2-cp311-cp311-win_amd64.whl; platform_system == "Windows" and python_version == "3.11"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.24+cpuavx2-cp310-cp310-win_amd64.whl; platform_system == "Windows" and python_version == "3.10"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.24+cpuavx2-cp39-cp39-win_amd64.whl; platform_system == "Windows" and python_version == "3.9"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.24+cpuavx2-cp38-cp38-win_amd64.whl; platform_system == "Windows" 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"
@@ -68,14 +68,14 @@ https://github.com/Dao-AILab/flash-attention/releases/download/v2.3.4/flash_attn
 https://github.com/Dao-AILab/flash-attention/releases/download/v2.3.4/flash_attn-2.3.4+cu122torch2.1cxx11abiFALSE-cp310-cp310-linux_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.10"
 https://github.com/Dao-AILab/flash-attention/releases/download/v2.3.4/flash_attn-2.3.4+cu122torch2.1cxx11abiFALSE-cp39-cp39-linux_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.9"
 https://github.com/Dao-AILab/flash-attention/releases/download/v2.3.4/flash_attn-2.3.4+cu122torch2.1cxx11abiFALSE-cp38-cp38-linux_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-0.2.23+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-0.2.23+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-0.2.23+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-0.2.23+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-0.2.23+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-0.2.23+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-0.2.23+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-0.2.23+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-0.2.24+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-0.2.24+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-0.2.24+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-0.2.24+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-0.2.24+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-0.2.24+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-0.2.24+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-0.2.24+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/GPTQ-for-LLaMa-CUDA/releases/download/0.1.1/gptq_for_llama-0.1.1+cu121-cp311-cp311-win_amd64.whl; platform_system == "Windows" and python_version == "3.11"
 https://github.com/jllllll/GPTQ-for-LLaMa-CUDA/releases/download/0.1.1/gptq_for_llama-0.1.1+cu121-cp310-cp310-win_amd64.whl; platform_system == "Windows" and python_version == "3.10"
 https://github.com/jllllll/GPTQ-for-LLaMa-CUDA/releases/download/0.1.1/gptq_for_llama-0.1.1+cu121-cp39-cp39-win_amd64.whl; platform_system == "Windows" and python_version == "3.9"