main
Update serve chat parsing (#46267)
ci: cover xet as well (runtime error) (#47338)
Fix multi token decode merging (#47762)
[fix] update integration test values (#47146)
[Fix] Fix multi-head latent attention (MLA) (#47761)
Fix `config.per_layer_config[layer_type]` lookup speed (#47539)
[Mistral] Add native tekken tokenizer support to AutoTokenizer (#47507)
update sonicmoe versions (#47769)
Use tiny Hub checkpoint in Qwen3ASR processor test (#47833)
Add GreedyLR adaptive learning rate scheduler (#44271)
[`peft`] Support key_mapping with PEFT models (#46766)
Deprecate the old response_schema (#47320)
Update AutoRound XPU/CPU backend (#47826)
CI: Log GitHub API diagnostics to stderr (#47635)
Remove deprecated `use_auth_token` parameter (#41666)
🚨 Pass tp_plan from lm_heads (#47253)
Fix path traversal via vocab-file arguments in tokenizer_config.json (#46279)
Use explicit nn.Buffer everywhere for modular (#47722)
docs(tests): fix typos in test comments (#47859)
GPU text generation: mMoved the encoded_prompt to correct device
🚨 [ALM] Add base model without head (#45534)
Add Granite-swa and Granitemoe-swa model support (#47179)
Better and more extensive tests for RoPE (#46912)
🚨 Delete duplicate code in backbone utils (#43323)
:rotating_light: Validate config attributes (#41250)
fix(flaky): enforce manual seed to reduce flakiness (#43794)
Use Python 3.9 syntax in tests (#37343)
PR CI with torch 2.13 (#47767)
Refactor image loading in tests to use load_test_image helper (#47218)
Raise `TypeError` instead of ValueError for invalid types (#38660)
🚨 Enable SDPA (and other attention backends) for T5 and propagate to the T5 family (#47014)
[Fix] Remove old automatic cross attn pattern from output recorders (#47117)
🚨 More V5 pipeline cleanup (#43325)
tests: reduce processor test memory usage and use tiny test assets (#47168)
rm slow tokenizers (#40936)
[tests] remove TF tests (uses of `require_tf`) (#38944)
Fix TP inference for tied embedding (#47503)
Fix models for which we don't have a dedicated tokenizer class, and the listed one is incorrect (#45936)
[MistralCommonBackend] Stop relying on mistral-common's Tokenized.text field (#47646)
incorrect model list update (#44880)
[style] Rework ruff rules and update all files (#43144)
[Tests] Fix slow video tensor creation from list of numpy arrays in SmolVLM (#44731)
audio tester class (#45391)