Models
Widgets replay real runs recorded ahead of time. Choose a tier to run real models in your browser instead: its models download from Hugging Face and stay in this browser until you remove them or it clears space.
Small
GPT-2 small (8-bit)
Predicts the next token.
- Size
- 128 MB
- Commit
- bf2c7f0
- Used in
- The guessing gameProbability and samplingGenerating textQuirks and limitsA prompt is the start of a documentReading both waysFrom predictor to assistantLearning from examples in the prompt
Checking…
all-MiniLM-L6-v2
Turns a sentence into a point.
- Size
- 24 MB
- Commit
- 751bff3
- Used in
- Meaning as position
Checking…
BERT base (8-bit)
Fills in a blank, reading both sides.
- Size
- 112 MB
- Commit
- ab680a3
- Used in
- Reading both ways
Checking…
CLIP ViT-B/32 text half (4-bit)
Turns a caption into a point it can compare with pictures.
- Size
- 128 MB
- Commit
- d15189d
- Used in
- Seeing pictures
Checking…
Large
Qwen2.5-0.5B-Instruct (4-bit)
Chats in its template, follows instructions, and calls tools.
- Size
- 793 MB
- Commit
- 22942cb
- Used in
- From predictor to assistantInstructions, roles, and the system promptLearning from examples in the promptTools
Checking…
ONNX Runtime
Runs every model.
- Size
- 27 MB
It comes from this site the first time a model runs, and your browser keeps it with the site's other files.