Qwen3.5-2B-mentria
Qwen3.5-2B in Q4 safetensors format for mentria-engine — a custom WebGPU runtime that runs the model entirely in your browser. Includes a Q4 vision tower for image input and a hot-swappable LoRA fine-tuned for motivational quotes.
This is the
middle tier of the mentria model ladder (
0.8B → 2B →
4B) — served to capable devices; lower-end devices fall back to the 0.8B tier.
Try it live:
- mentria.ai/tools/ai-chat — text + vision chat
- mentria.ai/tools/quote — quote generator (uses the bundled LoRA)
Files
| Path | Size | Purpose |
|---|
qwen3.5-2b-q4-tied.safetensors | ~1.1 GB | Language model — Q4_0 weights (MSE-optimal per-block scale) with tied embedding/lm_head (the embedding table is shared with the output projection, so it ships once). Loaded with allowTiedEmbed. |
qwen3.5-2b-vl-q4.safetensors | ~200 MB | Vision tower — Q4_0 patch embedding + transformer blocks |
tokenizer.json | ~12 MB | Qwen3.5 BPE tokenizer (248,320 vocab) |
tokenizer_config.json | ~16 KB | Special-token IDs, chat-template metadata |
chat_template.jinja | ~8 KB | Standalone chat template |
loras/quotes/adapter_config.json | ~400 B | LoRA manifest (peft_type: LORA, r=16, alpha=32) |
loras/quotes/adapter_model.safetensors | ~72 MB | LoRA weights — fine-tuned on the published quotes dataset |
Total cold-load on first visit: ~1.3 GB (LM + vision) or ~1.1 GB (LM only). Subsequent visits are instant — IndexedDB-cached.
Quantization quality
Measured against the BF16 base on a 129-task suite (24 VQA, 20 reasoning, 5 captions, 80 POPE-adversarial public VQA), greedy decoding @128 tokens. The Q4 weights are dequantized shader-exactly into the base layout, so the only difference is weight precision:
| Metric | BF16 base | This Q4 |
|---|
| VQA accuracy | 0.917 | 0.958 |
| Reasoning accuracy | 1.00 | 0.95 |
| POPE-adversarial accuracy | 0.863 | 0.875 |
| Caption F1 (lexical) | 0.653 | 0.544 |
Grade agreement with base: 0.96 across the 124 graded tasks — Q4 and BF16 give the same verdict on 96% of tasks.
Format note
These files target mentria-engine's specific Q4_0 safetensors layout. They are produced by an offline conversion pipeline from the upstream BF16 checkpoint (with MSE-optimal per-block Q4 scales) and are intended for use by mentria-engine specifically.
Model details
| |
|---|
| Base model | Qwen/Qwen3.5-2B |
| Architecture | Hybrid Gated-DeltaNet (18 layers) + GQA Attention (6 layers) + SwiGLU MLP, 24 layers, 2B params, hidden 2048 |
| Quantization | Q4_0 (group size 32, F16 scale per K-block, MSE-optimal scales) |
| Vision tower | 24-layer ViT, 1024 hidden, patch 16, spatial merge 2, projects to 2048-dim text embeddings |
| Tokenizer | Qwen3.5 BPE, 248,320 vocab |
| License | Apache 2.0 |
LoRA: loras/quotes/
Hot-swappable fine-tune for the motivational-quote use case.
| |
|---|
| Training data | mentriaai/motivational-quotes — 581 hand-curated original quotes, diversity-engineered |
| Hyperparameters | rank 16, alpha 32, dropout 0.05, LR 1.5e-4 cosine-decay (400 iters), AdamW, completion-format with prompt masking (no template/think tokens trained) |
| Target modules | mlp.{down,gate,up}_proj, self_attn.{q,k,v,o,a,b,g}_proj (full-coverage fine-tune) |
About mentria.ai
Mentria is a creative studio for tools, experiments, and visual transmissions. All tools run locally in your browser with zero server dependency.
