YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
Model Card: GenomeOcean-500M (FP8)
Generated: 2026-05-09T18:04:04-0700
Architecture
| Parameter | Value |
|---|---|
| Architecture | MistralForCausalLM |
| Model Type | mistral |
| Vocab Size | 4096 |
| Hidden Size | 1536 |
| Num Hidden Layers | 14 |
| Num Attention Heads | 8 |
| Intermediate Size | 6144 |
| Max Position Embeddings | 32768 |
| RoPE Theta | 1000000.0 |
Quantization Method
- Format: FP8 (E4M3) per-channel weight-only quantization
- Scale DType: float32 per-channel scales
- Method: Post-training quantization (PTQ) with per-channel E4M3 weights
Perplexity Results
| Metric | Value |
|---|---|
| Original PPL (BF16) | 39954.8889 |
| Quantized PPL (FP8) | 39753.4247 |
| PPL Difference | -201.4642 |
| PPL Difference (%) | -0.5% |
Quality Assessment: Excellent - negligible quality loss
Weight Fidelity
| Metric | Value |
|---|---|
| Mean Cosine Similarity | 0.999698 |
| Min Cosine Similarity | 0.999505 |
| Mean Relative L2 Error | 0.026571 |
| Max Relative L2 Error | 0.026663 |
| Layers Compared | 99 |
Compression
| Metric | Value |
|---|---|
| Original Size | 1.0825 GB |
| Quantized Size | 0.5361 GB |
| Compression Ratio | 49.52% |
| Space Saved | 0.55 GB |
Summary
The GenomeOcean-500M model was quantized from BF16 to FP8. Perplexity changed by -0.5% (original: 39954.8889, quantized: 39753.4247). Mean weight cosine similarity is 0.9997. Compression ratio is 49.52% (saved 0.55 GB).
- Downloads last month
- 14
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support