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:

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.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for mentriaai/Qwen3.5-2B-mentria

Finetuned
Qwen/Qwen3.5-2B
Quantized
(144)
this model