Commit History
Yesterday
-
Add a link to try the model in the LiteRT-LM Web chat demo
Verified signature484916c
5 days ago
-
Card: add a note on litert-lm's 10-minute send_message timeout
Verified signature7a5ec4f
6 days ago
-
litertlm_manifest.json: the externalized files' sha256, size and sections; min runtime 0.19.0 and the Android GPU known issue (LiteRT-LM#3986)
Verified signature228b45c -
Move the externalized files from the web branch to main (TFLiteWeights sections; Android GPU needs LiteRT-LM 0.19.0 or later)
Verified signaturea856703
11 days ago
-
Add a known runtime issue: a preface with tools and no messages fails to render the chat template
Verified signatureAdds the lines under Known runtime issues; nothing else in the card changes.076e556
16 days ago
-
Say that the CLI's Init counts engine creation twice (LiteRT-LM #4057)
Verified signatureThe CLI's printed Init includes both Init Total and the phases inside it. One sentence, no number changed.282baa6
29 days ago
-
manifest 0.1.3: capabilities.thinking.control = model (what the bundle's template does about thinking, derived from the template inside the bundle)
Verified signature7621a4f
30 days ago
-
Card: measured device / browser results (edge-compat, 2026-09)
Verified signature7689240
-
Card: the graph-rewrite reproduce command was missing the tool's required --report argument (text only, files unchanged)
Verified signature2905a31 -
GPU graph re-ship: KV-cache update and attention rewritten into the runtime's GPU composite ops plus a 16-token prefill signature; weights, tokenizer and metadata byte-identical; CPU output bit-identical (card and manifest updated)
Verified signatureaf26156
-
Chat template: accept the 0.18 content-parts form (string form unchanged); weights, tokenizer and executor metadata byte-identical
Verified signaturec29e765
-
Card: measured device / browser results (edge-compat, 2026-09)
Verified signaturef14a1ad
-
manifest: Android measured rows (Galaxy S26 GPU + same-device CPU control; Pixel 8a where run) and an Android recommendation where the evidence supports one
Verified signature543b513
-
Card: definition line (LiteRT, 2026-09)
Verified signatureeebc7d2
-
manifest: Galaxy S26 GPU rows from the S4 gate sweep (Android recommendation where none existed)
Verified signature77a37c0 -
Card: base_model_relation: quantized (list under the base model's Quantizations)
Verified signature9fbf9be
-
Add measured Raspberry Pi 5 (CPU) section
Verified signatureNumbers from the 2026-09 Raspberry Pi 5 LLM sweep: litert-lm benchmark 0.16.1, CPU backend, 4 threads, 256 prefill + 256 decode tokens, compile cache disabled, 3 invocations per file with cooldown, thermal throttle state verified 0x0 on every run, and a real-generation gate on the same backend before any number was recorded. Median + spread, never single-shot.64dea3f
-
Manifest: regenerate after the thought-channel re-pack (sha256 + capabilities.thinking)
Verified signature84b9900 -
Card: note the thought-channel declaration (metadata only, weights unchanged)
Verified signature8ac120b -
Declare the thought channel so the reasoning is separable and the thinking budget applies (metadata only; weights byte-identical)
Verified signature6a68e8e -
Declare the thought channel so the reasoning is separable and the thinking budget applies (metadata only; weights byte-identical)
Verified signaturecf297ef
-
Card: note the tokenizer-section fix (weights unchanged)
Verified signature49f281e -
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). litertlm_manifest.json regenerated in the same commit.
Verified signature2366ebb
-
Card: note the default-system-prompt fix (weights unchanged)
Verified signature9612ce2 -
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). litertlm_manifest.json regenerated in the same commit.
Verified signaturecf93e72 -
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). litertlm_manifest.json regenerated in the same commit.
Verified signature20a3a1f -
Card: correct the start_token note — measured numbers do not carry over
Verified signaturee44d6c0 -
Card: note the start_token fix (weights unchanged)
Verified signature0a8ad9d -
Regenerate litertlm_manifest.json for the new bundle sha256
Verified signatureThe bundle's metadata start_token was dropped, so its sha256 changed. Weights are byte-identical; the measured rows are unchanged.2551094 -
Drop the metadata start_token (weights unchanged)
Verified signatureThe metadata start_token held the literal string "None" (this tokenizer has no BOS), and the engine resolved it to a real vocabulary token - so every prompt began with the word "None", a stream the model was never trained on. Metadata-only change: every tokenizer and TFLite section is byte-identical to the previous file (verified by sha256 per section), so quality and speed are unchanged and the card's measured numbers still stand.b2ce10b
-
Add measured Galaxy S26 GPU backend section
Verified signatureFunctional verdict from the S26 GPU gate. Primary log: litertlm-convert/community_accel_work/s4_gpu_gate/results.jsonl. Delegation counts and peak RSS are read from that run's own log; throughput is deliberately absent, because an Android GPU speed is not quotable without a CPU row from the same phone and that control has not been run.5f8d413
-
Add litertlm_manifest.json — machine-readable deployment manifest (variant selection, backend recommendations, measured performance)
Verified signatureff4fc37
-
Add measured Pixel 8a OpenCL delegation (runtime capability; Gallery guidance unchanged)
Verified signature7c5ec04
-
Card: fill the iPhone prefill/TTFT from the run log (they were recorded; the previous note was wrong)
Verified signatureb9daee8 -
Card: measured Performance table (M4 Max CPU/GPU + on-device where measured), accuracy note
Verified signatureee27540
-
Add model (#1)
- Add model (b44fcea0db1fcd6a23d8e9776b0d6dbc25c173ed) Co-authored-by: Harsha Veena <hvt4@users.noreply.huggingface.co>a28ab40
-
Document Gallery 1.0.16 direct Hugging Face import + desktop LiteRT-LM CLI (serve/run)
Verified signaturea1498a4
-
docs: add Training data & PII section (submission-guideline compliance)
Verified signature21b9574
-
Upload folder using huggingface_hub
Verified signaturea2f0c54 -
initial commit
Verified signaturedb3081c