xinference 1.4.0__py3-none-any.whl → 1.5.0__py3-none-any.whl
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Potentially problematic release.
This version of xinference might be problematic. Click here for more details.
- xinference/_compat.py +1 -0
- xinference/_version.py +3 -3
- xinference/api/restful_api.py +54 -1
- xinference/client/restful/restful_client.py +82 -2
- xinference/constants.py +3 -0
- xinference/core/chat_interface.py +297 -83
- xinference/core/model.py +24 -3
- xinference/core/progress_tracker.py +16 -8
- xinference/core/supervisor.py +51 -1
- xinference/core/worker.py +315 -47
- xinference/deploy/cmdline.py +33 -1
- xinference/model/audio/core.py +11 -1
- xinference/model/audio/megatts.py +105 -0
- xinference/model/audio/model_spec.json +24 -1
- xinference/model/audio/model_spec_modelscope.json +26 -1
- xinference/model/core.py +14 -0
- xinference/model/embedding/core.py +6 -1
- xinference/model/flexible/core.py +6 -1
- xinference/model/image/core.py +6 -1
- xinference/model/image/model_spec.json +17 -1
- xinference/model/image/model_spec_modelscope.json +17 -1
- xinference/model/llm/__init__.py +4 -6
- xinference/model/llm/core.py +5 -0
- xinference/model/llm/llama_cpp/core.py +46 -17
- xinference/model/llm/llm_family.json +530 -85
- xinference/model/llm/llm_family.py +24 -1
- xinference/model/llm/llm_family_modelscope.json +572 -1
- xinference/model/llm/mlx/core.py +16 -2
- xinference/model/llm/reasoning_parser.py +3 -3
- xinference/model/llm/sglang/core.py +111 -13
- xinference/model/llm/transformers/__init__.py +14 -0
- xinference/model/llm/transformers/core.py +31 -6
- xinference/model/llm/transformers/deepseek_vl.py +1 -1
- xinference/model/llm/transformers/deepseek_vl2.py +287 -0
- xinference/model/llm/transformers/gemma3.py +17 -2
- xinference/model/llm/transformers/intern_vl.py +28 -18
- xinference/model/llm/transformers/minicpmv26.py +21 -2
- xinference/model/llm/transformers/qwen-omni.py +308 -0
- xinference/model/llm/transformers/qwen2_audio.py +1 -1
- xinference/model/llm/transformers/qwen2_vl.py +20 -4
- xinference/model/llm/utils.py +37 -15
- xinference/model/llm/vllm/core.py +184 -8
- xinference/model/llm/vllm/distributed_executor.py +320 -0
- xinference/model/rerank/core.py +22 -12
- xinference/model/utils.py +118 -1
- xinference/model/video/core.py +6 -1
- xinference/thirdparty/deepseek_vl2/__init__.py +31 -0
- xinference/thirdparty/deepseek_vl2/models/__init__.py +26 -0
- xinference/thirdparty/deepseek_vl2/models/configuration_deepseek.py +210 -0
- xinference/thirdparty/deepseek_vl2/models/conversation.py +310 -0
- xinference/thirdparty/deepseek_vl2/models/modeling_deepseek.py +1975 -0
- xinference/thirdparty/deepseek_vl2/models/modeling_deepseek_vl_v2.py +697 -0
- xinference/thirdparty/deepseek_vl2/models/processing_deepseek_vl_v2.py +675 -0
- xinference/thirdparty/deepseek_vl2/models/siglip_vit.py +661 -0
- xinference/thirdparty/deepseek_vl2/serve/__init__.py +0 -0
- xinference/thirdparty/deepseek_vl2/serve/app_modules/__init__.py +0 -0
- xinference/thirdparty/deepseek_vl2/serve/app_modules/gradio_utils.py +83 -0
- xinference/thirdparty/deepseek_vl2/serve/app_modules/overwrites.py +81 -0
- xinference/thirdparty/deepseek_vl2/serve/app_modules/presets.py +115 -0
- xinference/thirdparty/deepseek_vl2/serve/app_modules/utils.py +333 -0
- xinference/thirdparty/deepseek_vl2/serve/assets/Kelpy-Codos.js +100 -0
- xinference/thirdparty/deepseek_vl2/serve/assets/avatar.png +0 -0
- xinference/thirdparty/deepseek_vl2/serve/assets/custom.css +355 -0
- xinference/thirdparty/deepseek_vl2/serve/assets/custom.js +22 -0
- xinference/thirdparty/deepseek_vl2/serve/assets/favicon.ico +0 -0
- xinference/thirdparty/deepseek_vl2/serve/assets/simsun.ttc +0 -0
- xinference/thirdparty/deepseek_vl2/serve/inference.py +197 -0
- xinference/thirdparty/deepseek_vl2/utils/__init__.py +18 -0
- xinference/thirdparty/deepseek_vl2/utils/io.py +80 -0
- xinference/thirdparty/megatts3/__init__.py +0 -0
- xinference/thirdparty/megatts3/tts/frontend_function.py +175 -0
- xinference/thirdparty/megatts3/tts/gradio_api.py +93 -0
- xinference/thirdparty/megatts3/tts/infer_cli.py +277 -0
- xinference/thirdparty/megatts3/tts/modules/aligner/whisper_small.py +318 -0
- xinference/thirdparty/megatts3/tts/modules/ar_dur/ar_dur_predictor.py +362 -0
- xinference/thirdparty/megatts3/tts/modules/ar_dur/commons/layers.py +64 -0
- xinference/thirdparty/megatts3/tts/modules/ar_dur/commons/nar_tts_modules.py +73 -0
- xinference/thirdparty/megatts3/tts/modules/ar_dur/commons/rel_transformer.py +403 -0
- xinference/thirdparty/megatts3/tts/modules/ar_dur/commons/rot_transformer.py +649 -0
- xinference/thirdparty/megatts3/tts/modules/ar_dur/commons/seq_utils.py +342 -0
- xinference/thirdparty/megatts3/tts/modules/ar_dur/commons/transformer.py +767 -0
- xinference/thirdparty/megatts3/tts/modules/llm_dit/cfm.py +309 -0
- xinference/thirdparty/megatts3/tts/modules/llm_dit/dit.py +180 -0
- xinference/thirdparty/megatts3/tts/modules/llm_dit/time_embedding.py +44 -0
- xinference/thirdparty/megatts3/tts/modules/llm_dit/transformer.py +230 -0
- xinference/thirdparty/megatts3/tts/modules/wavvae/decoder/diag_gaussian.py +67 -0
- xinference/thirdparty/megatts3/tts/modules/wavvae/decoder/hifigan_modules.py +283 -0
- xinference/thirdparty/megatts3/tts/modules/wavvae/decoder/seanet_encoder.py +38 -0
- xinference/thirdparty/megatts3/tts/modules/wavvae/decoder/wavvae_v3.py +60 -0
- xinference/thirdparty/megatts3/tts/modules/wavvae/encoder/common_modules/conv.py +154 -0
- xinference/thirdparty/megatts3/tts/modules/wavvae/encoder/common_modules/lstm.py +51 -0
- xinference/thirdparty/megatts3/tts/modules/wavvae/encoder/common_modules/seanet.py +126 -0
- xinference/thirdparty/megatts3/tts/utils/audio_utils/align.py +36 -0
- xinference/thirdparty/megatts3/tts/utils/audio_utils/io.py +95 -0
- xinference/thirdparty/megatts3/tts/utils/audio_utils/plot.py +90 -0
- xinference/thirdparty/megatts3/tts/utils/commons/ckpt_utils.py +171 -0
- xinference/thirdparty/megatts3/tts/utils/commons/hparams.py +215 -0
- xinference/thirdparty/megatts3/tts/utils/text_utils/dict.json +1 -0
- xinference/thirdparty/megatts3/tts/utils/text_utils/ph_tone_convert.py +94 -0
- xinference/thirdparty/megatts3/tts/utils/text_utils/split_text.py +90 -0
- xinference/thirdparty/megatts3/tts/utils/text_utils/text_encoder.py +280 -0
- xinference/types.py +10 -0
- xinference/utils.py +54 -0
- xinference/web/ui/build/asset-manifest.json +6 -6
- xinference/web/ui/build/index.html +1 -1
- xinference/web/ui/build/static/css/main.0f6523be.css +2 -0
- xinference/web/ui/build/static/css/main.0f6523be.css.map +1 -0
- xinference/web/ui/build/static/js/main.58bd483c.js +3 -0
- xinference/web/ui/build/static/js/main.58bd483c.js.map +1 -0
- xinference/web/ui/node_modules/.cache/babel-loader/3bff8cbe9141f937f4d98879a9771b0f48e0e4e0dbee8e647adbfe23859e7048.json +1 -0
- xinference/web/ui/node_modules/.cache/babel-loader/4500b1a622a031011f0a291701e306b87e08cbc749c50e285103536b85b6a914.json +1 -0
- xinference/web/ui/node_modules/.cache/babel-loader/51709f5d3e53bcf19e613662ef9b91fb9174942c5518987a248348dd4e1e0e02.json +1 -0
- xinference/web/ui/node_modules/.cache/babel-loader/69081049f0c7447544b7cfd73dd13d8846c02fe5febe4d81587e95c89a412d5b.json +1 -0
- xinference/web/ui/node_modules/.cache/babel-loader/b8551e9775a01b28ae674125c688febe763732ea969ae344512e64ea01bf632e.json +1 -0
- xinference/web/ui/node_modules/.cache/babel-loader/bf2b211b0d1b6465eff512d64c869d748f803c5651a7c24e48de6ea3484a7bfe.json +1 -0
- xinference/web/ui/src/locales/en.json +2 -1
- xinference/web/ui/src/locales/zh.json +2 -1
- {xinference-1.4.0.dist-info → xinference-1.5.0.dist-info}/METADATA +128 -115
- {xinference-1.4.0.dist-info → xinference-1.5.0.dist-info}/RECORD +124 -63
- {xinference-1.4.0.dist-info → xinference-1.5.0.dist-info}/WHEEL +1 -1
- xinference/web/ui/build/static/css/main.b494ae7e.css +0 -2
- xinference/web/ui/build/static/css/main.b494ae7e.css.map +0 -1
- xinference/web/ui/build/static/js/main.3cea968e.js +0 -3
- xinference/web/ui/build/static/js/main.3cea968e.js.map +0 -1
- xinference/web/ui/node_modules/.cache/babel-loader/27bcada3ee8f89d21184b359f022fc965f350ffaca52c9814c29f1fc37121173.json +0 -1
- xinference/web/ui/node_modules/.cache/babel-loader/7f59e45e3f268ab8a4788b6fb024cf8dab088736dff22f5a3a39c122a83ab930.json +0 -1
- xinference/web/ui/node_modules/.cache/babel-loader/dcd60488509450bfff37bfff56de2c096d51de17dd00ec60d4db49c8b483ada1.json +0 -1
- xinference/web/ui/node_modules/.cache/babel-loader/e547bbb18abb4a474b675a8d5782d25617566bea0af8caa9b836ce5649e2250a.json +0 -1
- /xinference/web/ui/build/static/js/{main.3cea968e.js.LICENSE.txt → main.58bd483c.js.LICENSE.txt} +0 -0
- {xinference-1.4.0.dist-info → xinference-1.5.0.dist-info}/entry_points.txt +0 -0
- {xinference-1.4.0.dist-info → xinference-1.5.0.dist-info/licenses}/LICENSE +0 -0
- {xinference-1.4.0.dist-info → xinference-1.5.0.dist-info}/top_level.txt +0 -0
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"ast":null,"code":"import _slicedToArray from\"/home/runner/work/inference/inference/xinference/web/ui/node_modules/@babel/runtime/helpers/esm/slicedToArray.js\";import React,{createContext,useState}from'react';import{getEndpoint}from'./utils';import{jsx as _jsx}from\"react/jsx-runtime\";export var ApiContext=/*#__PURE__*/createContext();export var ApiContextProvider=function ApiContextProvider(_ref){var children=_ref.children;var _useState=useState(false),_useState2=_slicedToArray(_useState,2),isCallingApi=_useState2[0],setIsCallingApi=_useState2[1];var _useState3=useState(false),_useState4=_slicedToArray(_useState3,2),isUpdatingModel=_useState4[0],setIsUpdatingModel=_useState4[1];var _useState5=useState(''),_useState6=_slicedToArray(_useState5,2),errorMsg=_useState6[0],setErrorMsg=_useState6[1];var _useState7=useState(''),_useState8=_slicedToArray(_useState7,2),successMsg=_useState8[0],setSuccessMsg=_useState8[1];var endPoint=getEndpoint();return/*#__PURE__*/_jsx(ApiContext.Provider,{value:{isCallingApi:isCallingApi,setIsCallingApi:setIsCallingApi,isUpdatingModel:isUpdatingModel,setIsUpdatingModel:setIsUpdatingModel,endPoint:endPoint,errorMsg:errorMsg,setErrorMsg:setErrorMsg,successMsg:successMsg,setSuccessMsg:setSuccessMsg},children:children});};","map":{"version":3,"names":["React","createContext","useState","getEndpoint","jsx","_jsx","ApiContext","ApiContextProvider","_ref","children","_useState","_useState2","_slicedToArray","isCallingApi","setIsCallingApi","_useState3","_useState4","isUpdatingModel","setIsUpdatingModel","_useState5","_useState6","errorMsg","setErrorMsg","_useState7","_useState8","successMsg","setSuccessMsg","endPoint","Provider","value"],"sources":["/home/runner/work/inference/inference/xinference/web/ui/src/components/apiContext.js"],"sourcesContent":["import React, { createContext, useState } from 'react'\n\nimport { getEndpoint } from './utils'\n\nexport const ApiContext = createContext()\n\nexport const ApiContextProvider = ({ children }) => {\n const [isCallingApi, setIsCallingApi] = useState(false)\n const [isUpdatingModel, setIsUpdatingModel] = useState(false)\n const [errorMsg, setErrorMsg] = useState('')\n const [successMsg, setSuccessMsg] = useState('')\n const endPoint = getEndpoint()\n\n return (\n <ApiContext.Provider\n value={{\n isCallingApi,\n setIsCallingApi,\n isUpdatingModel,\n setIsUpdatingModel,\n endPoint,\n errorMsg,\n setErrorMsg,\n successMsg,\n setSuccessMsg,\n }}\n >\n {children}\n </ApiContext.Provider>\n )\n}\n"],"mappings":"6IAAA,MAAO,CAAAA,KAAK,EAAIC,aAAa,CAAEC,QAAQ,KAAQ,OAAO,CAEtD,OAASC,WAAW,KAAQ,SAAS,QAAAC,GAAA,IAAAC,IAAA,yBAErC,MAAO,IAAM,CAAAC,UAAU,cAAGL,aAAa,CAAC,CAAC,CAEzC,MAAO,IAAM,CAAAM,kBAAkB,CAAG,QAArB,CAAAA,kBAAkBA,CAAAC,IAAA,CAAqB,IAAf,CAAAC,QAAQ,CAAAD,IAAA,CAARC,QAAQ,CAC3C,IAAAC,SAAA,CAAwCR,QAAQ,CAAC,KAAK,CAAC,CAAAS,UAAA,CAAAC,cAAA,CAAAF,SAAA,IAAhDG,YAAY,CAAAF,UAAA,IAAEG,eAAe,CAAAH,UAAA,IACpC,IAAAI,UAAA,CAA8Cb,QAAQ,CAAC,KAAK,CAAC,CAAAc,UAAA,CAAAJ,cAAA,CAAAG,UAAA,IAAtDE,eAAe,CAAAD,UAAA,IAAEE,kBAAkB,CAAAF,UAAA,IAC1C,IAAAG,UAAA,CAAgCjB,QAAQ,CAAC,EAAE,CAAC,CAAAkB,UAAA,CAAAR,cAAA,CAAAO,UAAA,IAArCE,QAAQ,CAAAD,UAAA,IAAEE,WAAW,CAAAF,UAAA,IAC5B,IAAAG,UAAA,CAAoCrB,QAAQ,CAAC,EAAE,CAAC,CAAAsB,UAAA,CAAAZ,cAAA,CAAAW,UAAA,IAAzCE,UAAU,CAAAD,UAAA,IAAEE,aAAa,CAAAF,UAAA,IAChC,GAAM,CAAAG,QAAQ,CAAGxB,WAAW,CAAC,CAAC,CAE9B,mBACEE,IAAA,CAACC,UAAU,CAACsB,QAAQ,EAClBC,KAAK,CAAE,CACLhB,YAAY,CAAZA,YAAY,CACZC,eAAe,CAAfA,eAAe,CACfG,eAAe,CAAfA,eAAe,CACfC,kBAAkB,CAAlBA,kBAAkB,CAClBS,QAAQ,CAARA,QAAQ,CACRN,QAAQ,CAARA,QAAQ,CACRC,WAAW,CAAXA,WAAW,CACXG,UAAU,CAAVA,UAAU,CACVC,aAAa,CAAbA,aACF,CAAE,CAAAjB,QAAA,CAEDA,QAAQ,CACU,CAAC,CAE1B,CAAC","ignoreList":[]},"metadata":{},"sourceType":"module","externalDependencies":[]}
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"ast":null,"code":"import Snackbar from'@mui/material/Snackbar';import React,{useContext}from'react';import{Alert}from'./alertComponent';import{ApiContext}from'./apiContext';import{jsx as _jsx}from\"react/jsx-runtime\";var SuccessMessageSnackBar=function SuccessMessageSnackBar(){var _useContext=useContext(ApiContext),successMsg=_useContext.successMsg,setSuccessMsg=_useContext.setSuccessMsg;var handleClose=function handleClose(event,reason){if(reason==='clickaway'){return;}setSuccessMsg('');};return/*#__PURE__*/_jsx(Snackbar,{open:successMsg!=='',autoHideDuration:3000,anchorOrigin:{vertical:'top',horizontal:'center'},onClose:handleClose,children:/*#__PURE__*/_jsx(Alert,{severity:\"success\",onClose:handleClose,sx:{width:'100%'},children:successMsg})});};export default SuccessMessageSnackBar;","map":{"version":3,"names":["Snackbar","React","useContext","Alert","ApiContext","jsx","_jsx","SuccessMessageSnackBar","_useContext","successMsg","setSuccessMsg","handleClose","event","reason","open","autoHideDuration","anchorOrigin","vertical","horizontal","onClose","children","severity","sx","width"],"sources":["/home/runner/work/inference/inference/xinference/web/ui/src/components/successMessageSnackBar.js"],"sourcesContent":["import Snackbar from '@mui/material/Snackbar'\nimport React, { useContext } from 'react'\n\nimport { Alert } from './alertComponent'\nimport { ApiContext } from './apiContext'\n\nconst SuccessMessageSnackBar = () => {\n const { successMsg, setSuccessMsg } = useContext(ApiContext)\n\n const handleClose = (event, reason) => {\n if (reason === 'clickaway') {\n return\n }\n setSuccessMsg('')\n }\n\n return (\n <Snackbar\n open={successMsg !== ''}\n autoHideDuration={3000}\n anchorOrigin={{ vertical: 'top', horizontal: 'center' }}\n onClose={handleClose}\n >\n <Alert severity=\"success\" onClose={handleClose} sx={{ width: '100%' }}>\n {successMsg}\n </Alert>\n </Snackbar>\n )\n}\n\nexport default SuccessMessageSnackBar\n"],"mappings":"AAAA,MAAO,CAAAA,QAAQ,KAAM,wBAAwB,CAC7C,MAAO,CAAAC,KAAK,EAAIC,UAAU,KAAQ,OAAO,CAEzC,OAASC,KAAK,KAAQ,kBAAkB,CACxC,OAASC,UAAU,KAAQ,cAAc,QAAAC,GAAA,IAAAC,IAAA,yBAEzC,GAAM,CAAAC,sBAAsB,CAAG,QAAzB,CAAAA,sBAAsBA,CAAA,CAAS,CACnC,IAAAC,WAAA,CAAsCN,UAAU,CAACE,UAAU,CAAC,CAApDK,UAAU,CAAAD,WAAA,CAAVC,UAAU,CAAEC,aAAa,CAAAF,WAAA,CAAbE,aAAa,CAEjC,GAAM,CAAAC,WAAW,CAAG,QAAd,CAAAA,WAAWA,CAAIC,KAAK,CAAEC,MAAM,CAAK,CACrC,GAAIA,MAAM,GAAK,WAAW,CAAE,CAC1B,OACF,CACAH,aAAa,CAAC,EAAE,CAAC,CACnB,CAAC,CAED,mBACEJ,IAAA,CAACN,QAAQ,EACPc,IAAI,CAAEL,UAAU,GAAK,EAAG,CACxBM,gBAAgB,CAAE,IAAK,CACvBC,YAAY,CAAE,CAAEC,QAAQ,CAAE,KAAK,CAAEC,UAAU,CAAE,QAAS,CAAE,CACxDC,OAAO,CAAER,WAAY,CAAAS,QAAA,cAErBd,IAAA,CAACH,KAAK,EAACkB,QAAQ,CAAC,SAAS,CAACF,OAAO,CAAER,WAAY,CAACW,EAAE,CAAE,CAAEC,KAAK,CAAE,MAAO,CAAE,CAAAH,QAAA,CACnEX,UAAU,CACN,CAAC,CACA,CAAC,CAEf,CAAC,CAED,cAAe,CAAAF,sBAAsB","ignoreList":[]},"metadata":{},"sourceType":"module","externalDependencies":[]}
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"ast":null,"code":"export var llmAllDataKey=['model_uid','model_name','model_type','model_engine','model_format','model_size_in_billions','quantization','n_worker','n_gpu','n_gpu_layers','replica','request_limits','worker_ip','gpu_idx','download_hub','model_path','reasoning_content','gguf_quantization','gguf_model_path','cpu_offload','peft_model_config'];export var additionalParameterTipList={'transformers':['torch_dtype','device'],'llama.cpp':['n_ctx','use_mmap','use_mlock'],'vllm':['block_size','gpu_memory_utilization','max_num_seqs','max_model_len','guided_decoding_backend','scheduling_policy','tensor_parallel_size','pipeline_parallel_size','enable_prefix_caching','enable_chunked_prefill','enforce_eager','cpu_offload_gb','disable_custom_all_reduce','limit_mm_per_prompt','model_quantization','mm_processor_kwargs','min_pixels','max_pixels'],'sglang':['mem_fraction_static','attention_reduce_in_fp32','tp_size','dp_size','chunked_prefill_size','cpu_offload_gb','enable_dp_attention','enable_ep_moe'],'mlx':['cache_limit_gb','max_kv_size']};export var featureModels=[{type:'llm',feature_models:['deepseek-v3','deepseek-r1','deepseek-r1-distill-qwen','deepseek-r1-distill-llama','qwen2.5-instruct','qwen2.5-vl-instruct','qwen2.5-coder-instruct','QwQ-32B','llama-3.1-instruct','gemma-3-it']},{type:'embedding',feature_models:['bge-large-zh-v1.5','bge-large-en-v1.5','bge-m3','gte-Qwen2','jina-embeddings-v3']},{type:'rerank',feature_models:[]},{type:'image',feature_models:['FLUX.1-dev','FLUX.1-schnell','sd3.5-large','HunyuanDiT-v1.2','sd3.5-medium']},{type:'audio',feature_models:['CosyVoice2-0.5B','FishSpeech-1.5','F5-TTS','ChatTTS','SenseVoiceSmall','whisper-large-v3']},{type:'video',feature_models:[]}];","map":{"version":3,"names":["llmAllDataKey","additionalParameterTipList","featureModels","type","feature_models"],"sources":["/home/runner/work/inference/inference/xinference/web/ui/src/scenes/launch_model/data/data.js"],"sourcesContent":["export const llmAllDataKey = [\n 'model_uid',\n 'model_name',\n 'model_type',\n 'model_engine',\n 'model_format',\n 'model_size_in_billions',\n 'quantization',\n 'n_worker',\n 'n_gpu',\n 'n_gpu_layers',\n 'replica',\n 'request_limits',\n 'worker_ip',\n 'gpu_idx',\n 'download_hub',\n 'model_path',\n 'reasoning_content',\n 'gguf_quantization',\n 'gguf_model_path',\n 'cpu_offload',\n 'peft_model_config',\n]\n\nexport const additionalParameterTipList = {\n 'transformers': ['torch_dtype', 'device'],\n 'llama.cpp': ['n_ctx', 'use_mmap', 'use_mlock'],\n 'vllm': [\n 'block_size',\n 'gpu_memory_utilization',\n 'max_num_seqs',\n 'max_model_len',\n 'guided_decoding_backend',\n 'scheduling_policy',\n 'tensor_parallel_size',\n 'pipeline_parallel_size',\n 'enable_prefix_caching',\n 'enable_chunked_prefill',\n 'enforce_eager',\n 'cpu_offload_gb',\n 'disable_custom_all_reduce',\n 'limit_mm_per_prompt',\n 'model_quantization',\n 'mm_processor_kwargs',\n 'min_pixels',\n 'max_pixels',\n ],\n 'sglang': [\n 'mem_fraction_static',\n 'attention_reduce_in_fp32',\n 'tp_size',\n 'dp_size',\n 'chunked_prefill_size',\n 'cpu_offload_gb',\n 'enable_dp_attention',\n 'enable_ep_moe',\n ],\n 'mlx': ['cache_limit_gb', 'max_kv_size'],\n}\n\nexport const featureModels = [\n {\n type: 'llm',\n feature_models: [\n 'deepseek-v3',\n 'deepseek-r1',\n 'deepseek-r1-distill-qwen',\n 'deepseek-r1-distill-llama',\n 'qwen2.5-instruct',\n 'qwen2.5-vl-instruct',\n 'qwen2.5-coder-instruct',\n 'QwQ-32B',\n 'llama-3.1-instruct',\n 'gemma-3-it',\n ],\n },\n {\n type: 'embedding',\n feature_models: [\n 'bge-large-zh-v1.5',\n 'bge-large-en-v1.5',\n 'bge-m3',\n 'gte-Qwen2',\n 'jina-embeddings-v3',\n ],\n },\n {\n type: 'rerank',\n feature_models: [],\n },\n {\n type: 'image',\n feature_models: [\n 'FLUX.1-dev',\n 'FLUX.1-schnell',\n 'sd3.5-large',\n 'HunyuanDiT-v1.2',\n 'sd3.5-medium',\n ],\n },\n {\n type: 'audio',\n feature_models: [\n 'CosyVoice2-0.5B',\n 'FishSpeech-1.5',\n 'F5-TTS',\n 'ChatTTS',\n 'SenseVoiceSmall',\n 'whisper-large-v3',\n ],\n },\n {\n type: 'video',\n feature_models: [],\n },\n]\n"],"mappings":"AAAA,MAAO,IAAM,CAAAA,aAAa,CAAG,CAC3B,WAAW,CACX,YAAY,CACZ,YAAY,CACZ,cAAc,CACd,cAAc,CACd,wBAAwB,CACxB,cAAc,CACd,UAAU,CACV,OAAO,CACP,cAAc,CACd,SAAS,CACT,gBAAgB,CAChB,WAAW,CACX,SAAS,CACT,cAAc,CACd,YAAY,CACZ,mBAAmB,CACnB,mBAAmB,CACnB,iBAAiB,CACjB,aAAa,CACb,mBAAmB,CACpB,CAED,MAAO,IAAM,CAAAC,0BAA0B,CAAG,CACxC,cAAc,CAAE,CAAC,aAAa,CAAE,QAAQ,CAAC,CACzC,WAAW,CAAE,CAAC,OAAO,CAAE,UAAU,CAAE,WAAW,CAAC,CAC/C,MAAM,CAAE,CACN,YAAY,CACZ,wBAAwB,CACxB,cAAc,CACd,eAAe,CACf,yBAAyB,CACzB,mBAAmB,CACnB,sBAAsB,CACtB,wBAAwB,CACxB,uBAAuB,CACvB,wBAAwB,CACxB,eAAe,CACf,gBAAgB,CAChB,2BAA2B,CAC3B,qBAAqB,CACrB,oBAAoB,CACpB,qBAAqB,CACrB,YAAY,CACZ,YAAY,CACb,CACD,QAAQ,CAAE,CACR,qBAAqB,CACrB,0BAA0B,CAC1B,SAAS,CACT,SAAS,CACT,sBAAsB,CACtB,gBAAgB,CAChB,qBAAqB,CACrB,eAAe,CAChB,CACD,KAAK,CAAE,CAAC,gBAAgB,CAAE,aAAa,CACzC,CAAC,CAED,MAAO,IAAM,CAAAC,aAAa,CAAG,CAC3B,CACEC,IAAI,CAAE,KAAK,CACXC,cAAc,CAAE,CACd,aAAa,CACb,aAAa,CACb,0BAA0B,CAC1B,2BAA2B,CAC3B,kBAAkB,CAClB,qBAAqB,CACrB,wBAAwB,CACxB,SAAS,CACT,oBAAoB,CACpB,YAAY,CAEhB,CAAC,CACD,CACED,IAAI,CAAE,WAAW,CACjBC,cAAc,CAAE,CACd,mBAAmB,CACnB,mBAAmB,CACnB,QAAQ,CACR,WAAW,CACX,oBAAoB,CAExB,CAAC,CACD,CACED,IAAI,CAAE,QAAQ,CACdC,cAAc,CAAE,EAClB,CAAC,CACD,CACED,IAAI,CAAE,OAAO,CACbC,cAAc,CAAE,CACd,YAAY,CACZ,gBAAgB,CAChB,aAAa,CACb,iBAAiB,CACjB,cAAc,CAElB,CAAC,CACD,CACED,IAAI,CAAE,OAAO,CACbC,cAAc,CAAE,CACd,iBAAiB,CACjB,gBAAgB,CAChB,QAAQ,CACR,SAAS,CACT,iBAAiB,CACjB,kBAAkB,CAEtB,CAAC,CACD,CACED,IAAI,CAAE,OAAO,CACbC,cAAc,CAAE,EAClB,CAAC,CACF","ignoreList":[]},"metadata":{},"sourceType":"module","externalDependencies":[]}
|