SupraTTS-1.0-Preview
🎙
5
1.03M-parameter English TTS running in your browser
🏗️ Building on HF
AxionLab
AxionLab-official
AI & ML interests
Owner of SupraLabs
Recent Activity
liked a Space about 3 hours ago
SupraLabs/supratts-1-0-preview new activity about 3 hours ago
SupraLabs/README:Fix README and revert top the new fixed version liked a model about 15 hours ago
SupraLabs/SupraTTS-1.0-PreviewOrganizations
liked a
Space about 3 hours ago
Fix README and revert top the new fixed version
1
#13 opened about 4 hours ago
by 
DedeProGames
liked a
model about 15 hours ago
Fixed "Arthurblg1802" and updated the models
4
#10 opened about 18 hours ago
by DedeProGames
New SUPREME Organization Card
10
#9 opened about 19 hours ago
by DedeProGames
updated a
Space about 19 hours ago
liked a
Space 1 day ago
Update index.html
#1 opened 1 day ago
by AxionLab-official
liked a
model 1 day ago
Update index.html
#229 opened 1 day ago
by AxionLab-official
updated a
model 1 day ago
reacted to DedeProGames's post with 🧠🚀🔥 1 day ago
Post
2,748
Meet Supra2.5-Ultra-Preview-001. 230M parameters, ~40.4B training tokens.
SupraLabs has released the first preview of its Supra2.5-Ultra family: a compact, Llama-style base model for text completion and small-model experimentation.
What's inside?
• 230M parameters
• 32 transformer layers
• 2,048-token context window
• Grouped Query Attention: 12 attention heads, 4 KV heads
• 49,152-token vocabulary
• SwiGLU activation, RoPE and tied embeddings
• BF16 weights and Hugging Face Transformers support
• Apache 2.0 license
Training
The model was trained on approximately 40.4B tokens from FineWeb-Edu, DCLM-Baseline and FineMath using a single A100 SXM.
Reported zero-shot results
• HellaSwag: 37.18% normalized accuracy
• PIQA: 65.78% accuracy
• ARC-Easy: 57.11% accuracy
• ARC-Challenge: 26.28% normalized accuracy
Want to experiment?
Code on REPO.
Model, weights and usage example:
SupraLabs/Supra2.5-Ultra-Preview-001
SupraLabs has released the first preview of its Supra2.5-Ultra family: a compact, Llama-style base model for text completion and small-model experimentation.
What's inside?
• 230M parameters
• 32 transformer layers
• 2,048-token context window
• Grouped Query Attention: 12 attention heads, 4 KV heads
• 49,152-token vocabulary
• SwiGLU activation, RoPE and tied embeddings
• BF16 weights and Hugging Face Transformers support
• Apache 2.0 license
Training
The model was trained on approximately 40.4B tokens from FineWeb-Edu, DCLM-Baseline and FineMath using a single A100 SXM.
Reported zero-shot results
• HellaSwag: 37.18% normalized accuracy
• PIQA: 65.78% accuracy
• ARC-Easy: 57.11% accuracy
• ARC-Challenge: 26.28% normalized accuracy
Want to experiment?
Code on REPO.
Model, weights and usage example:
SupraLabs/Supra2.5-Ultra-Preview-001
- 2 replies
SupraLabs/Supra2.5-Ultra-Preview-001
🤯👍 2
#12988 opened 2 days ago
by AxionLab-official
Update index.html
1
#228 opened 2 days ago
by AxionLab-official
Why Llama?
2
#1 opened 2 days ago
by 
qikp
reacted to mithulaartigala's post with 🔥 2 days ago
liked 2
datasets 3 days ago