Hugging Face's logo

litert-community
/
North-Micro-Vision-Instruct

Image-Text-to-Text
LiteRT-LM
LiteRT
litertlm
on-device
edge
vlm
multimodal
vision-language
cohere
Model card Files
xet
 Community
North-Micro-Vision-Instruct

Commit History

5 days ago

  1. litertlm_manifest.json: the 2026-10-15 re-packed North-Micro-Vision-Instruct_int4.litertlm and North-Micro-Vision-Instruct_wi8.litertlm (tokenizer section replaced by the upstream tokenizer: sha256 and tokenizer section size) and the digit known_issues items

    mlboydaisuke
    Verified signature
    0a4f513

  2. Card: 2026-10-15 tokenizer fix (tokenizer section replaced by the upstream tokenizer): 17 times 23 is read as 17 × 23; the tokenizer, 8-question gate and device lines now name the files they describe

    mlboydaisuke
    Verified signature
    201a8a0

  3. Re-pack North-Micro-Vision-Instruct_wi8.litertlm: tokenizer section replaced by the upstream tokenizer (tokenizer.json of CohereLabs/North-Micro-Vision-Instruct at fa548d9a). The earlier file carried the tokenizer.json that transformers' CohereTokenizer class writes (NFC + per-digit split), so 17 reached the model as 1, 7 and was read as 1.7. Every other section, the LlmMetadata included, is byte-identical.

    mlboydaisuke
    Verified signature
    d812a8a

  4. Re-pack North-Micro-Vision-Instruct_int4.litertlm: tokenizer section replaced by the upstream tokenizer (tokenizer.json of CohereLabs/North-Micro-Vision-Instruct at fa548d9a). The earlier file carried the tokenizer.json that transformers' CohereTokenizer class writes (NFC + per-digit split), so 17 reached the model as 1, 7 and was read as 1.7. Every other section, the LlmMetadata included, is byte-identical.

    mlboydaisuke
    Verified signature
    c97bb1e

12 days ago

  1. manifest 0.1.3: capabilities.thinking.control = never (what the bundle's template does about thinking, derived from the template inside the bundle)

    mlboydaisuke
    Verified signature
    54df699

  1. Card: measured device / browser results (edge-compat, 2026-09)

    mlboydaisuke
    Verified signature
    dcc2656

  1. Add litertlm_manifest.json (deployment manifest 0.1.2: variant selection, verified backends, Android/Apple recommendations, measured rows with conditions)

    mlboydaisuke
    Verified signature
    6f4ad08

  2. Card: base_model_relation: quantized (list under the base model's Quantizations)

    mlboydaisuke
    Verified signature
    4fbdc98

  1. Card: note the chat-template fix for role-'assistant' history (weights unchanged)

    mlboydaisuke
    Verified signature
    b85ff50

  2. Chat template fix (metadata-only, weights unchanged): the template matched history turns only on role 'model', so a conversation created through the Conversation API with the standard role 'assistant' (e.g. create_conversation(messages=[...]) restoring a chat) silently dropped every assistant turn from the rendered prompt. The role conditions now also accept 'assistant'. Every other section is byte-identical; single-turn renders, token ids, prefill and greedy answers are unchanged (verified on the LiteRT-LM runtime). See google-ai-edge/LiteRT-LM#3417.

    mlboydaisuke
    Verified signature
    4f0b2f5

  1. Add measured Galaxy S26 GPU backend section⁠

    mlboydaisuke
    Verified signature
    5b8ffed

  1. card: on-device demo table (Pixel 8a answers on two HF sample photos)

    mlboydaisuke
    Verified signature
    b6ee5ed

  2. card: iPhone 17 Pro re-gate on the published file (vision probes 2/2, peak 2.9-3.0 GB, TTFT 2.3-5.0 s) + Core AI same-device paragraph

    mlboydaisuke
    Verified signature
    6378e25

  3. int4 blockwise-32 decoder variant + int8 vision

    mlboydaisuke
    Verified signature
    4d418e4

  4. wi8 decoder (fp32_fp16 hint) + int8 vision, static-512 fast_vlm

    mlboydaisuke
    Verified signature
    7a27ff0

  5. North-Micro-Vision-Instruct LiteRT-LM card

    mlboydaisuke
    Verified signature
    d658443

  6. initial commit

    mlboydaisuke
    Verified signature
    7e99c2d