oobabooga/text-generation-webui
1
0
Fork 0
mirror of https://github.com/oobabooga/text-generation-webui.git synced 2026-04-04 22:27:29 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

llama.cpp: Add speculative decoding (#6891)

Browse source
This commit is contained in:
oobabooga 2025-04-23 20:10:16 -03:00 • committed by GitHub
parent 9424ba17c8
commit e99c20bcb0
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
7 changed files with 61 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

11
modules/ui_model_menu.py
View file

@ -92,6 +92,17 @@ def create_ui():
shared.gradio['exllamav2_info'] = gr.Markdown("ExLlamav2_HF is recommended over ExLlamav2 for better integration with extensions and more consistent sampling behavior across loaders.")
shared.gradio['tensorrt_llm_info'] = gr.Markdown('* TensorRT-LLM has to be installed manually in a separate Python 3.10 environment at the moment. For a guide, consult the description of [this PR](https://github.com/oobabooga/text-generation-webui/pull/5715). \n\n* `max_seq_len` is only used when `cpp-runner` is checked.\n\n* `cpp_runner` does not support streaming at the moment.')
# Speculative decoding
with gr.Accordion("Speculative decoding", open=False, elem_classes='tgw-accordion') as shared.gradio['speculative_decoding_accordion']:
with gr.Row():
shared.gradio['model_draft'] = gr.Dropdown(label="model-draft", choices=utils.get_available_models(), value=lambda: shared.draft_model_name, elem_classes='slim-dropdown', interactive=not mu)
ui.create_refresh_button(shared.gradio['model_draft'], lambda: None, lambda: {'choices': utils.get_available_models()}, 'refresh-button', interactive=not mu)
shared.gradio['draft_max'] = gr.Number(label="draft-max", precision=0, step=1, value=shared.args.draft_max, info='Number of tokens to draft for speculative decoding.')
shared.gradio['gpu_layers_draft'] = gr.Slider(label="gpu-layers-draft", minimum=0, maximum=256, value=shared.args.gpu_layers_draft, info='Number of layers to offload to the GPU for the draft model.')
shared.gradio['device_draft'] = gr.Textbox(label="device-draft", value=shared.args.device_draft, info='Comma-separated list of devices to use for offloading the draft model.')
shared.gradio['ctx_size_draft'] = gr.Number(label="ctx-size-draft", precision=0, step=256, value=shared.args.ctx_size_draft, info='Size of the prompt context for the draft model. If 0, uses the same as the main model.')
with gr.Column():
with gr.Row():
shared.gradio['autoload_model'] = gr.Checkbox(value=shared.settings['autoload_model'], label='Autoload the model', info='Whether to load the model as soon as it is selected in the Model dropdown.', interactive=not mu)