MichaelWelsch/gemma-4-E2B-it-litert-community-128k-mtp

Gemma 4 E2B LiteRT-LM 128k + MTP

Experimental .litertlm bundle for Gemma 4 E2B derived from the LiteRT community artifact and patched for:

• max_num_tokens: 131072
• MTP / speculative decoding support retained
• native LiteRT-LM runtime compatibility

Status

What is verified on host 100.96.1.7:

• short-prompt inference works on CPU
• speculative decoding works
• LiteRT-LM logs show target_number=131072
• long-context prefill above 32k starts correctly without falling back to 32k

What is not claimed yet:

• full production qualification
• GPU qualification
• parity with all Gemma4 sizes
• Qwen qualification

Artifact

• file: model.litertlm
• sha256: 274e5c461e754cbd05423bab734e7765d9757443fb5591edb7aba6f9f186550a
• size_bytes: 2584805376

Notes

This is an experimental artifact upload for CTOX integration work. Use with caution.