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

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

Card: compatibility line (edge-compat, 2026-09)

manifest: same-device CPU control rows on the Galaxy S26 for the flagship GPU recommendation (S7 backfill 2026-09-05); the recommendation is rewritten from the measured pair (measured win, kept for prefill, or reversed to cpu)

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

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

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

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.

Card: note the tokenizer-section fix (weights unchanged)

Replace the tokenizer section with the upstream tokenizer.json (weights, graph and metadata unchanged). The previous SentencePiece conversion of the BPE tokenizer encoded standalone accented/special characters to the wrong ids, turned characters without a vocabulary entry (emoji, some accented capitals) into the end-of-text/end-of-turn token, and could not match the token it had reused as UNK; prompts now tokenize identically to the upstream tokenizer (verified on the LiteRT-LM runtime, see the card note).

Card: note the default-system-prompt fix (weights unchanged)

Restore the vendor's default system prompt in the chat template (weights unchanged). The converter's template probe never renders the block the upstream template emits when the caller sends no system message, so the bundle silently dropped it; with no system message the prompt now renders byte-identical to the upstream chat template's output (see the card note).

Card: correct the start_token note — measured numbers do not carry over

Card: note the start_token fix (weights unchanged)

Drop the metadata start_token (weights unchanged)

Add measured Galaxy S26 GPU backend section

Card: state where each on-device figure actually comes from; replace the iPhone row with a fresh measurement where one was taken

Card: verified one-command usage, measured Performance table (backends gated on real generation), accuracy note

Document Gallery 1.0.16 direct Hugging Face import + desktop LiteRT-LM CLI (serve/run)

Upload README.md with huggingface_hub

Upload README.md with huggingface_hub

Upload InternVL3-2B.litertlm with huggingface_hub

Upload README.md with huggingface_hub

Upload README.md with huggingface_hub

Upload InternVL3-2B.litertlm with huggingface_hub

Upload InternVL3-2B.litertlm with huggingface_hub

Upload README.md with huggingface_hub

initial commit