litert-community/Qwen3.5-4B

Chat template: accept the 0.18 content-parts form (string form unchanged); weights, tokenizer and executor metadata byte-identical

Card: definition line (LiteRT, 2026-09)

manifest: Galaxy S26 GPU rows from the S4 gate sweep (Android recommendation where none existed)

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

Add measured Raspberry Pi 5 (CPU) section

manifest: Pixel 8a measured row

Mixed INT4: measured Pixel 8a (8 GB) CPU row

manifest: add Mixed INT4 variant

Qwen3.5-4B Mixed INT4 (int4 b32 linears, int8 embed/lm_head, fp32act)

Add Mixed INT4 variant (LiteRT-LM #1658)

Add litertlm_manifest.json — machine-readable deployment manifest (variant selection, backend recommendations, measured performance)

Card: state where GPU execution is verified, and scope the Android decode note

Qwen3.5-4B int8 (fp32act, 7-signature ladder)

Qwen3.5-4B LiteRT-LM card

initial commit