Spaces:
Running
Running
Commit History
-
Restore cancel generation feature with improved UI integration
a94befb -
Remove cancel generation feature as it didn't work
efa3ae6 -
Restore proper cancel functionality
- Revert stopping criteria to check cancel_event.is_set() - Confirmed stopping criteria mechanism works (generation stopped after 1 token in test) - Cancel should now work properly, stopping generation after current token when clickedafa1066 -
Test stopping criteria effectiveness
- Temporarily modify CancelStoppingCriteria to always return True - This will test if stopping criteria are being checked during generation - If generation stops after first token, stopping criteria work; if not, they are ignoredf0d2ab7 -
Improve cancel generation responsiveness
- Add additional cancel checks before each yield in the streaming loop - Ensures generation stops immediately when cancel is triggered, preventing partial responses from being displayed - Complements the StoppingCriteria to provide more robust cancellation48e50b8 -
Fix cancel generation to gracefully stop ongoing response generation
- Add StoppingCriteria import and CancelStoppingCriteria class - Integrate stopping criteria into generation pipeline to halt token generation when cancel event is set - Ensures generation stops at the model level, preventing unnecessary GPU usagea73d8f4 -
remove qwen3 30b-a3b and qwen3 next 80b-a3b
4af617b -
Remove AOT indication from UI duration estimate
f97cbfc -
Remove AOT compilation completely and enable use_cache
426163f -
Fix dynamic_shapes kwargs to match inputs structure for AOT compilation
273acf8 -
Fix AOT compilation dynamic_shapes to match expected arg names for torch.export.export
0a99dfc -
Update requirements.txt
Verified signature4d2c362 -
Update requirements.txt
Verified signature48d806a -
Add dynamic GPU time estimate indicator to UI
- Shows estimated GPU seconds for current inference settings - Updates in real-time when model, search settings, or max tokens change - Displays model size, AOT status, and search status - Helps users manage precious ZeroGPU time budget effectivelyfc989b4 -
Improve model size detection: replace ad-hoc string parsing with reliable params_b field in MODELS dict
ab92e0d -
Add qwen 80b-a3b
8cdf3e1 -
Set better defaults for free-tier users: Qwen3-1.7B model, 1024 max tokens, search disabled
2cae073 -
Adjust duration estimation for H200 performance - reduce conservative estimates
de766da -
Use actual parameter count for AOT decision instead of string matching
e3e334f -
Make AOT compilation conditional for models >= 2B parameters to optimize free tier usage
4500f92 -
Add AOT compilation optimization for ZeroGPU acceleration
a7866ff -
add 4 20b+ models after enabling dynamic gpu duration
Verified signaturefea2910 -
Add dynamic duration calculation for ZeroGPU acceleration
6073cc2
-
make qwen-4b default
Verified signatured3726c6 -
disable two models that cannot run or too run too slowly on hf spaces with zerogpu
3dc7ced -
Update app.py
Verified signaturef1fa55c -
Update app.py
Verified signaturef07d6ab -
Update app.py
Verified signature0992852 -
add original apriel 15b
Verified signature2b25033 -
use apriel 8bit
Verified signaturea4681bd -
run Apriel on 4bit
Verified signature2cadf8a -
Add Apriel-1.5-15b-Thinker
Verified signature3665b54
-
Update app.py
Verified signature7f654b2 -
Update app.py
Verified signature15b78c7 -
Update requirements.txt
Verified signature4addcb2 -
Update requirements.txt
Verified signature49c0c5a -
Update requirements.txt
Verified signature831c9e8
-
Update app.py
Verified signature7356fa6 -
add 4 models from qwen3 family
Verified signature048cfc4 -
add qwen3 32b awq
Verified signatureb9efb74 -
Update app.py
Verified signature5e03586 -
Update README.md
Verified signaturec6e816c -
Update requirements.txt
Verified signatureff4ad1d -
Update app.py
Verified signaturee5a1663 -
Update app.py
Verified signaturede64679 -
Update app.py
Verified signature4418827 -
Update app.py
Verified signaturef2f4310 -
Update app.py
Verified signature42db70b -
feat(models): add Granite-4.0-Micro and Qwen3-4B-Instruct-2507 to MODELS registry
Verified signatureAdded Granite-4.0-Micro (3B dense instruct model from IBM) under ~3B parameter category Added Qwen3-4B-Instruct-2507 (updated non-thinking instruct variant with 4B params and 256K context) under 4B–4.3B category Included detailed descriptions based on official Hugging Face model cards Maintained consistent formatting and category structurec30a7f7 -
feat(models): Added three new models
Verified signatureServiceNow-AI/Apriel-1.5-15b-Thinker (15B, multimodal reasoning) ServiceNow-AI/Apriel-5B-Instruct (5B, instruction-tuned) ai21labs/AI21-Jamba-Reasoning-3B (3B, hybrid Transformer–Mamba with 256K context) Reordered entire MODELS dictionary from largest (~15B) to smallest (~135M) Grouped models into logical size-based sections with updated descriptions using official Hugging Face metadata Enhanced model descriptions with key capabilities, benchmarks, and usage context from source model cards3c22497