Parallax Files

Highlights

• Add docker image for dgx spark
• Support huggingface offline
• Refactor executor and p2p server use sub-process instead of sub-thread
• Add paged attn naive kernel

What's Changed

• chore(lattica): upgrade lattica to 1.0.10 by @gufengc in https://github.com/GradientHQ/parallax/pull/221
• fix(executor): initialize http_requests list for ingesting new requests by @RWL-Dittrich in https://github.com/GradientHQ/parallax/pull/217
• feat(backend): add flags use hugging face cache by @JasonOE in https://github.com/GradientHQ/parallax/pull/220
• fix(backend): add param_hosting_ratio for estimate varm required by @sl-gn in https://github.com/GradientHQ/parallax/pull/224
• feat(scheduler): Launch TP>0 as subprocesses & fit scheduler by @TianyiZhao1437 in https://github.com/GradientHQ/parallax/pull/222
• fix(bug): Prevent node duplication during leftover node allocation by @IAMDAVID0920 in https://github.com/GradientHQ/parallax/pull/209
• fix(backend): fix monkey patch taking no effect by @TianyiZhao1437 in https://github.com/GradientHQ/parallax/pull/228
• chore(lattica): show lattica log in debug mode by @gufengc in https://github.com/GradientHQ/parallax/pull/231
• feat(scheduler): warmup before global rebalance by @JasonOE in https://github.com/GradientHQ/parallax/pull/204
• fix(bug): just change kvcache and param_host name by @wasamtc in https://github.com/GradientHQ/parallax/pull/229
• fix(scheduler): ignore node update if it is not in node list by @gufengc in https://github.com/GradientHQ/parallax/pull/232
• docs(readme): add partner zai by @sl-gn in https://github.com/GradientHQ/parallax/pull/233
• docs(fqa): add faq in user guide by @sl-gn in https://github.com/GradientHQ/parallax/pull/234
• docs(faq): add two qa by @sl-gn in https://github.com/GradientHQ/parallax/pull/235
• fix(test): fix executor test on gpu by @TianyiZhao1437 in https://github.com/GradientHQ/parallax/pull/236
• fix(server): surface chat-template errors by @Odysseusailoon in https://github.com/GradientHQ/parallax/pull/208
• fix(backend): fix vram estimate error when use hf local cache by @sl-gn in https://github.com/GradientHQ/parallax/pull/238
• fix(backend): skip load model info if model path is speficied by @gufengc in https://github.com/GradientHQ/parallax/pull/241
• chore(lattica): upgrade lattica to 1.0.12 by @gufengc in https://github.com/GradientHQ/parallax/pull/243
• feat(pre-commit): Hide whitespace by @yuhao-zh in https://github.com/GradientHQ/parallax/pull/249
• fix(backend): fix model list api error if not init model info from config by @sl-gn in https://github.com/GradientHQ/parallax/pull/250
• docs(dgx spark): Add dgx spark docker by @gufengc in https://github.com/GradientHQ/parallax/pull/251
• feat(backend): add vllm support by @yuhao-zh in https://github.com/GradientHQ/parallax/pull/223
• feat(docker): build docker for dgx spark by @gufengc in https://github.com/GradientHQ/parallax/pull/244
• chore(docker): build docker daily by @gufengc in https://github.com/GradientHQ/parallax/pull/253
• fix(tokenizer): fix tokenizer bug by @gufengc in https://github.com/GradientHQ/parallax/pull/255
• fix(eos): fix eos check by @gufengc in https://github.com/GradientHQ/parallax/pull/256
• chore(sglang): upgrade sglang to 0.5.5 by @gufengc in https://github.com/GradientHQ/parallax/pull/257
• feat(frontend): supports display gpu number in node by @xz-gradient in https://github.com/GradientHQ/parallax/pull/258
• chore(docs): update docker doc by @gufengc in https://github.com/GradientHQ/parallax/pull/259
• fix(model_download): exit when node's network cannot connect huggingface.co by @yuhao-zh in https://github.com/GradientHQ/parallax/pull/252
• fix(model): fix kimi k2 on mac by @gufengc in https://github.com/GradientHQ/parallax/pull/260
• fix(tokenizer): fix eos token id if it is a list by @gufengc in https://github.com/GradientHQ/parallax/pull/263
• fix(loader): fix bug when tie_word_embeddings is true by @yuhao-zh in https://github.com/GradientHQ/parallax/pull/264
• chore(lattica): upgrade lattica to 1.0.13 by @gufengc in https://github.com/GradientHQ/parallax/pull/268
• fix(scheduler): async get non-stream response to avoid block other re… by @gufengc in https://github.com/GradientHQ/parallax/pull/269
• feat(eos): add ignore_eos parameter by @gufengc in https://github.com/GradientHQ/parallax/pull/270
• feat(model): test model acc by @yuhao-zh in https://github.com/GradientHQ/parallax/pull/262
• fix(chat): fix bug when call chat api with non-stream by @sl-gn in https://github.com/GradientHQ/parallax/pull/274
• feat(kernel): add paged attn naive kernel by @yuhao-zh in https://github.com/GradientHQ/parallax/pull/273
• feat(backend): add sglang lora params of gpu by @wasamtc in https://github.com/GradientHQ/parallax/pull/272
• fix(api): fix wrong field in api response by @sl-gn in https://github.com/GradientHQ/parallax/pull/279
• feat(backend): subprocess for executor and gradient server by @JasonOE in https://github.com/GradientHQ/parallax/pull/254
• fix(dependency): use 4.57.1 transformers and add linux test by @gufengc in https://github.com/GradientHQ/parallax/pull/280
• chore(version): update version to 0.1.1 by @sl-gn in https://github.com/GradientHQ/parallax/pull/282

New Contributors

• @RWL-Dittrich made their first contribution in https://github.com/GradientHQ/parallax/pull/217
• @wasamtc made their first contribution in https://github.com/GradientHQ/parallax/pull/229
• @Odysseusailoon made their first contribution in https://github.com/GradientHQ/parallax/pull/208

Full Changelog: https://github.com/GradientHQ/parallax/compare/v0.1.0...v0.1.1