Releases: Aleph-Alpha/pharia-inference-sdk
Releases · Aleph-Alpha/pharia-inference-sdk
Release list
v0.1.6
v0.1.5
v0.1.4
v0.1.3
v0.1.2
0.1.2 (2025-06-11)
Features
FileInstructionFinetuningDataRepository(17babff)- add
context_size-property toAlephAlphaModel(#820) (ae7e634) - Add
eot_tokenproperty toControlModeland derived classes. (#906) (90291eb) - Add function to upload to trace viewer to Tracer. (#870) (a571531)
- Add TextControl for Instruct and Chat models (#1104) (ceb1756)
- harden model initialization (#884) (6777c00)
- Increased ruff checks for better code quality(#908) (d6c0311)
- Introduce
InstructionFinetuningDataRepository(#1033) (17babff) - Introduce an unified trace format for exporting (#844) (f765d27)
- make ChatModels instruct capable (#1064) (f24a6c4)
- Replace langdetect with lingua (#897) (a4d9be4)
- restructure tests according to src structure (#872) (d1509fa)
- Studio integration of Trace Submission PHS-616 (#975) (ff47799)
- Use urllib.urljoin where appropriate (#986) (8772f88)
Bug Fixes
ExpandChunksis fast even for large documents (#842) (30b9441)- allow negative min score in SearchQuery (#1184) (709a414)
- lru_cache memory consumption (#907) (30d1724)
- make highlight test stable (#916) (20d2bd5)
- notebook test stability, PromptBasedClassify with Llama (#1212) (c967183)
- remove llama3.0 from models and defaults (#1054) (24618cf)
- SingleChunkQa crashing on specific prompt ordering (#931) (dd801ff)
- Throw a warning instead of an error incase a non recommended model is selected for a ControlMode. (#892) (7aab1d7)
- tokenizer with no whitespace prefix will no longer fail (#1195) (96b0d47)
- update names of pharia-1 models to lowercase (#1091) (b7621d5)
- use "utf-8" encoding when reading FileTracer traces (#1241) (456dec0)