mirror of
https://github.com/oobabooga/text-generation-webui.git
synced 2026-04-04 22:27:29 +00:00
New llama.cpp loader (#6846)
This commit is contained in:
oobabooga
GitHub
parent
5c2f8d828e
commit
ae54d8faaa
23 changed files with 471 additions and 999 deletions
|
|
@ -87,7 +87,7 @@ def create_ui():
|
|||
shared.gradio['n_gpu_layers'] = gr.Slider(label="n-gpu-layers", minimum=0, maximum=256, value=shared.args.n_gpu_layers, info='Must be greater than 0 for the GPU to be used. ⚠️ Lower this value if you can\'t load the model.')
|
||||
shared.gradio['threads'] = gr.Slider(label="threads", minimum=0, step=1, maximum=256, value=shared.args.threads)
|
||||
shared.gradio['threads_batch'] = gr.Slider(label="threads_batch", minimum=0, step=1, maximum=256, value=shared.args.threads_batch)
|
||||
shared.gradio['n_batch'] = gr.Slider(label="n_batch", minimum=1, maximum=2048, step=1, value=shared.args.n_batch)
|
||||
shared.gradio['batch_size'] = gr.Slider(label="batch_size", minimum=1, maximum=4096, step=1, value=shared.args.batch_size)
|
||||
shared.gradio['hqq_backend'] = gr.Dropdown(label="hqq_backend", choices=["PYTORCH", "PYTORCH_COMPILE", "ATEN"], value=shared.args.hqq_backend)
|
||||
shared.gradio['n_ctx'] = gr.Number(label="n_ctx", precision=0, step=256, value=shared.args.n_ctx, info='Context length. ⚠️ Lower this value if you can\'t load the model. Common values: 2048, 4096, 8192, 16384, 32768.')
|
||||
shared.gradio['max_seq_len'] = gr.Number(label='max_seq_len', precision=0, step=256, value=shared.args.max_seq_len, info='Context length. ⚠️ Lower this value if you can\'t load the model. Common values: 2048, 4096, 8192, 16384, 32768.')
|
||||
|
|
@ -99,22 +99,19 @@ def create_ui():
|
|||
shared.gradio['compress_pos_emb'] = gr.Number(label='compress_pos_emb', value=shared.args.compress_pos_emb, precision=2, info='Positional embeddings compression factor. Should be set to (context length) / (model\'s original context length). Equal to 1/rope_freq_scale.')
|
||||
shared.gradio['compute_dtype'] = gr.Dropdown(label="compute_dtype", choices=["bfloat16", "float16", "float32"], value=shared.args.compute_dtype, info='Used by load-in-4bit.')
|
||||
shared.gradio['quant_type'] = gr.Dropdown(label="quant_type", choices=["nf4", "fp4"], value=shared.args.quant_type, info='Used by load-in-4bit.')
|
||||
shared.gradio['attention_sink_size'] = gr.Number(label="attention_sink_size", value=shared.args.attention_sink_size, precision=0, info='StreamingLLM: number of sink tokens. Only used if the trimmed prompt doesn\'t share a prefix with the old prompt.')
|
||||
shared.gradio['num_experts_per_token'] = gr.Number(label="Number of experts per token", value=shared.args.num_experts_per_token, info='Only applies to MoE models like Mixtral.')
|
||||
|
||||
with gr.Column():
|
||||
shared.gradio['tensorcores'] = gr.Checkbox(label="tensorcores", value=shared.args.tensorcores, info='NVIDIA only: use llama-cpp-python compiled without GGML_CUDA_FORCE_MMQ. This may improve performance on newer cards.')
|
||||
shared.gradio['load_in_8bit'] = gr.Checkbox(label="load-in-8bit", value=shared.args.load_in_8bit)
|
||||
shared.gradio['load_in_4bit'] = gr.Checkbox(label="load-in-4bit", value=shared.args.load_in_4bit)
|
||||
shared.gradio['torch_compile'] = gr.Checkbox(label="torch-compile", value=shared.args.torch_compile, info='Compile the model with torch.compile for improved performance.')
|
||||
shared.gradio['flash_attn'] = gr.Checkbox(label="flash_attn", value=shared.args.flash_attn, info='Use flash-attention.')
|
||||
shared.gradio['use_flash_attention_2'] = gr.Checkbox(label="use_flash_attention_2", value=shared.args.use_flash_attention_2, info='Set use_flash_attention_2=True while loading the model.')
|
||||
shared.gradio['streaming_llm'] = gr.Checkbox(label="streaming_llm", value=shared.args.streaming_llm, info='(experimental) Activate StreamingLLM to avoid re-evaluating the entire prompt when old messages are removed.')
|
||||
shared.gradio['auto_devices'] = gr.Checkbox(label="auto-devices", value=shared.args.auto_devices)
|
||||
shared.gradio['cpu'] = gr.Checkbox(label="cpu", value=shared.args.cpu, info='llama.cpp: Use llama-cpp-python compiled without GPU acceleration. Transformers: use PyTorch in CPU mode.')
|
||||
shared.gradio['disk'] = gr.Checkbox(label="disk", value=shared.args.disk)
|
||||
shared.gradio['row_split'] = gr.Checkbox(label="row_split", value=shared.args.row_split, info='Split the model by rows across GPUs. This may improve multi-gpu performance.')
|
||||
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['no_kv_offload'] = gr.Checkbox(label="no_kv_offload", value=shared.args.no_kv_offload, info='Do not offload the K, Q, V to the GPU. This saves VRAM but reduces the performance.')
|
||||
shared.gradio['no_mul_mat_q'] = gr.Checkbox(label="no_mul_mat_q", value=shared.args.no_mul_mat_q, info='Disable the mulmat kernels.')
|
||||
shared.gradio['no_mmap'] = gr.Checkbox(label="no-mmap", value=shared.args.no_mmap)
|
||||
shared.gradio['mlock'] = gr.Checkbox(label="mlock", value=shared.args.mlock)
|
||||
|
|
@ -129,7 +126,6 @@ def create_ui():
|
|||
shared.gradio['no_sdpa'] = gr.Checkbox(label="no_sdpa", value=shared.args.no_sdpa)
|
||||
shared.gradio['cfg_cache'] = gr.Checkbox(label="cfg-cache", value=shared.args.cfg_cache, info='Necessary to use CFG with this loader.')
|
||||
shared.gradio['cpp_runner'] = gr.Checkbox(label="cpp-runner", value=shared.args.cpp_runner, info='Enable inference with ModelRunnerCpp, which is faster than the default ModelRunner.')
|
||||
shared.gradio['logits_all'] = gr.Checkbox(label="logits_all", value=shared.args.logits_all, info='Needs to be set for perplexity evaluation to work with this loader. Otherwise, ignore it, as it makes prompt processing slower.')
|
||||
shared.gradio['trust_remote_code'] = gr.Checkbox(label="trust-remote-code", value=shared.args.trust_remote_code, info='Set trust_remote_code=True while loading the tokenizer/model. To enable this option, start the web UI with the --trust-remote-code flag.', interactive=shared.args.trust_remote_code)
|
||||
shared.gradio['no_use_fast'] = gr.Checkbox(label="no_use_fast", value=shared.args.no_use_fast, info='Set use_fast=False while loading the tokenizer.')
|
||||
shared.gradio['llamacpp_HF_info'] = gr.Markdown("llamacpp_HF loads llama.cpp as a Transformers model. To use it, you need to place your GGUF in a subfolder of models/ with the necessary tokenizer files.\n\nYou can use the \"llamacpp_HF creator\" menu to do that automatically.")
|
||||
|
|
@ -147,15 +143,6 @@ def create_ui():
|
|||
shared.gradio['download_model_button'] = gr.Button("Download", variant='primary', interactive=not mu)
|
||||
shared.gradio['get_file_list'] = gr.Button("Get file list", interactive=not mu)
|
||||
|
||||
with gr.Tab("llamacpp_HF creator"):
|
||||
with gr.Row():
|
||||
shared.gradio['gguf_menu'] = gr.Dropdown(choices=utils.get_available_ggufs(), value=lambda: shared.model_name, label='Choose your GGUF', elem_classes='slim-dropdown', interactive=not mu)
|
||||
ui.create_refresh_button(shared.gradio['gguf_menu'], lambda: None, lambda: {'choices': utils.get_available_ggufs()}, 'refresh-button', interactive=not mu)
|
||||
|
||||
shared.gradio['unquantized_url'] = gr.Textbox(label="Enter the URL for the original (unquantized) model", info="Example: https://huggingface.co/lmsys/vicuna-13b-v1.5", max_lines=1)
|
||||
shared.gradio['create_llamacpp_hf_button'] = gr.Button("Submit", variant="primary", interactive=not mu)
|
||||
gr.Markdown("This will move your gguf file into a subfolder of `models` along with the necessary tokenizer files.")
|
||||
|
||||
with gr.Tab("Customize instruction template"):
|
||||
with gr.Row():
|
||||
shared.gradio['customized_template'] = gr.Dropdown(choices=utils.get_available_instruction_templates(), value='None', label='Select the desired instruction template', elem_classes='slim-dropdown')
|
||||
|
|
@ -195,7 +182,6 @@ def create_event_handlers():
|
|||
shared.gradio['download_model_button'].click(download_model_wrapper, gradio('custom_model_menu', 'download_specific_file'), gradio('model_status'), show_progress=True)
|
||||
shared.gradio['get_file_list'].click(partial(download_model_wrapper, return_links=True), gradio('custom_model_menu', 'download_specific_file'), gradio('model_status'), show_progress=True)
|
||||
shared.gradio['autoload_model'].change(lambda x: gr.update(visible=not x), gradio('autoload_model'), gradio('load_model'))
|
||||
shared.gradio['create_llamacpp_hf_button'].click(create_llamacpp_hf, gradio('gguf_menu', 'unquantized_url'), gradio('model_status'), show_progress=True)
|
||||
shared.gradio['customized_template_submit'].click(save_instruction_template, gradio('model_menu', 'customized_template'), gradio('model_status'), show_progress=True)
|
||||
|
||||
|
||||
|
|
@ -286,34 +272,11 @@ def download_model_wrapper(repo_id, specific_file, progress=gr.Progress(), retur
|
|||
yield traceback.format_exc().replace('\n', '\n\n')
|
||||
|
||||
|
||||
def create_llamacpp_hf(gguf_name, unquantized_url, progress=gr.Progress()):
|
||||
try:
|
||||
downloader = importlib.import_module("download-model").ModelDownloader()
|
||||
|
||||
progress(0.0)
|
||||
model, branch = downloader.sanitize_model_and_branch_names(unquantized_url, None)
|
||||
|
||||
yield ("Getting the tokenizer files links from Hugging Face")
|
||||
links, sha256, is_lora, is_llamacpp = downloader.get_download_links_from_huggingface(model, branch, text_only=True)
|
||||
output_folder = Path(shared.args.model_dir) / (re.sub(r'(?i)\.gguf$', '', gguf_name) + "-HF")
|
||||
|
||||
yield (f"Downloading tokenizer to `{output_folder}/`")
|
||||
downloader.download_model_files(model, branch, links, sha256, output_folder, progress_bar=progress, threads=4, is_llamacpp=False)
|
||||
|
||||
# Move the GGUF
|
||||
(Path(shared.args.model_dir) / gguf_name).rename(output_folder / gguf_name)
|
||||
|
||||
yield (f"Model saved to `{output_folder}/`.\n\nYou can now load it using llamacpp_HF.")
|
||||
except:
|
||||
progress(1.0)
|
||||
yield traceback.format_exc().replace('\n', '\n\n')
|
||||
|
||||
|
||||
def update_truncation_length(current_length, state):
|
||||
if 'loader' in state:
|
||||
if state['loader'].lower().startswith('exllama'):
|
||||
return state['max_seq_len']
|
||||
elif state['loader'] in ['llama.cpp', 'llamacpp_HF']:
|
||||
elif state['loader'] == 'llama.cpp':
|
||||
return state['n_ctx']
|
||||
|
||||
return current_length
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue