Model Card: GenomeOcean-100M (GGUF_Q8_0)
Generated: 2026-05-09T20:10:10-0700
Architecture
| Parameter | Value |
|---|
| Architecture | MistralForCausalLM |
| Model Type | mistral |
| Vocab Size | 4096 |
| Hidden Size | 768 |
| Num Hidden Layers | 12 |
| Num Attention Heads | 8 |
| Intermediate Size | 3072 |
| Max Position Embeddings | 32768 |
| RoPE Theta | 1000000.0 |
Quantization Method
- Format: GGUF Q8_0 (8-bit integer quantization)
- Method: Post-training quantization via llama.cpp
Perplexity Results
| Metric | Value |
|---|
| Original PPL (BF16) | 40804.858 |
| Quantized PPL (GGUF_Q8_0) | 40818.7321 |
| PPL Difference | 13.8741 |
| PPL Difference (%) | 0.03% |
Quality Assessment: Excellent - negligible quality loss
Weight Fidelity
| Metric | Value |
|---|
| Mean Cosine Similarity | 0.999976 |
| Min Cosine Similarity | 0.999909 |
| Mean Relative L2 Error | 0.004286 |
| Max Relative L2 Error | 0.008096 |
| Layers Compared | 111 |
Compression
| Metric | Value |
|---|
| Original Size | 0.2394 GB |
| Quantized Size | 0.1302 GB |
| Compression Ratio | 54.38% |
| Space Saved | 0.11 GB |
Summary
The GenomeOcean-100M model was quantized from BF16 to GGUF_Q8_0.
Perplexity changed by 0.03% (original: 40804.858, quantized: 40818.7321).
Mean weight cosine similarity is 1.0000.
Compression ratio is 54.38% (saved 0.11 GB).
