Problems
From the last year
When using Gemma 4's thinking mode (
enable_thinking=True) with a max_tokens budget in the range of 512–1024, the model sometimes returns a response containing only the <channel|> delimiter and npython gemma thinking-mode inference token-budget 102 tokens
After deploying Gemma 4 E4B for inference, throughput plateaus at approximately 9-10 tokens/second regardless of serving framework. Switching between vLLM, SGLang, and Unsloth produces identical ceili
python gemma inference throughput vllm 69 tokens
When training Gemma 4 (4B or 31B variants) using HuggingFace's
DPOTrainer with text-only prompt/chosen/rejected triples, training fails immediately with:python gemma huggingface trl dpo-trainer 114 tokens