What's Changed
- Introduce model configs to abstract pairings of models and hardware by @yk in #2194
- Add recent changes to eval_model/manual/sampling_report.py by @andreaskoepf in #2191
- fix: ghcr.io build for mulitplatform. includes Apple silicon by @melvinebenezer in #2151
- Revert "fix: ghcr.io build for mulitplatform. includes Apple silicon" by @andreaskoepf in #2199
- Update CODEOWNERS for website by @AbdBarho in #2200
- Post llama merge fixes by @andreaskoepf in #2188
- feature : Alpaca dataset by @theblackcat102 in #2205
- Russian Translation Updated + Stuff by @0x22almostEvil in #2197
- Instruction Dataset: Retrieval-based grounded model generated Q-A pairs (BART version) by @michaelthwan in #2170
- Get available auth providers from inference server by @AbdBarho in #2207
- fixes text client to work with new debug login workflow by @yk in #2212
- Fix horizontal scrolling on mobile by @AbdBarho in #2211
- Update Ukrainian translation by @nmeln in #2214
- Use new inference model config / API by @AbdBarho in #2208
- Added CORS origins to inference settings by @yk in #2217
- Adjusted deployment notebooks for inference by @yk in #2213
- Add ability for inference backend to revoke auth refresh tokens by @olliestanley in #2175
- Fix to rank_datasets.py by @olliestanley in #2220
- Fixed bugs in deployment notebook (Sorry 🙃) by @yk in #2219
- Reduce star motion by @AbdBarho in #2215
- Add Inference sign out functionality by @AbdBarho in #2218
- SFT Rejection Sampling using RM by @shahules786 in #2225
- Various improvements to the dev setup by @yk in #2228
- update deps by @notmd in #2227
- Style updates to chat UI by @AbdBarho in #2226
- Add migrations to web db by @AbdBarho in #2233
- Create worker metrics manually for more control by @yk in #2229
- Sending MessageRead along with error to client by @yk in #2230
- Enabling Threads and Retry for Web Chat by @yk in #2232
- Export script: Fix duplicate loading of models by @andreaskoepf in #2231
- Add re-rank cli utility by @andreaskoepf in #2243
- Provide minimal documentation of oasst-data module and file format by @andreaskoepf in #2237
- update warning and improve readme in model training by @CloseChoice in #2246
- Improved worker script and documentation thereof by @yk in #2247
- Correlation metrics for Reward Model by @shahules786 in #2251
- Revert "Correlation metrics for Reward Model" by @andreaskoepf in #2253
- Add simple OIG data loader by @andreaskoepf in #2260
- Add correlation metrics for Reward Modeling by @shahules786 in #2266
- Expose env vars globally by @notmd in #2244
- Add Esperanto Language [fixed] by @0x22almostEvil in #2271
- Use LLaMA impl of Huggingface Transformers by @andreaskoepf in #2263
- Fix GPTNeoX-20B training by @dvruette in #2240
- Updated Turkish language by @irfantogluk in #2270
- Add loader for CodeAlpaca-20k & gpt4all_pruned dataset by @andreaskoepf in #2273
- Add support for Cerebras-GPT for training by @olliestanley in #2276
- typo in parsing openai/summarize_from_feedback by @mikegarts in #2268
- Add rng_seed parameter to trainers by @andreaskoepf in #2254
- Computing message queue positions by @yk in #2235
- Remove assigning eos token id (llama compatibility) by @andreaskoepf in #2280
- Fix call-to-action responsiveness by @theopfr in #2290
- Added max size to work queue and an error response if full when enqueuing by @yk in #2279
- Fix loading of Nebulous/gpt4all_pruned dataset by @andreaskoepf in #2291
- Move create chat button to the top by @AbdBarho in #2292
- remove CUDA_VISIBLE_DEVICES= which is user specific by @kno10 in #2295
- Use trusted clients for inference auth by @AbdBarho in #2278
- Add missing variables to deployment job by @AbdBarho in #2297
- two more datasets by @mikegarts in #2301
- Added a link to chat to the sidebar by @yk in #2303
- Added CTA buttons to the frontpage by @yk in #2302
- changed basic hf server to support quantization and streaming by @yk in #2293
New Contributors
- @michaelthwan made their first contribution in #2170
- @CloseChoice made their first contribution in #2246
- @irfantogluk made their first contribution in #2270
- @mikegarts made their first contribution in #2268
- @kno10 made their first contribution in #2295
Full Changelog: v0.0.1-beta57...v0.0.1-beta58