Issues

• append_tokens_to_cache commits canvas K/V with padded RoPE positions after sample_next_canvas switched to unpadded positions

  google-deepmind/gemma#808 · Filter by author Varun-51 opened on Oct 1, 2026

• gemma4:26b concurrent decode loses EOS (length at num_predict); qwen3.8-27b IDENTICAL dual on same GB10/Ollama harness is 3/3 clean #18442

  google-deepmind/gemma#798 · Filter by author scottmcmorrow opened on Sep 14, 2026

• Proposal: publish Gemma conformance test vectors

  google-deepmind/gemma#786 · Filter by author sylvesterkaczmarek opened on Aug 31, 2026

• Measurement: Gemma 4 and instruction-injection via third-party documentation

  google-deepmind/gemma#784 · Filter by author bpappin opened on Aug 26, 2026

• Gemma4 12B year mistake issue

  google-deepmind/gemma#778 · Filter by author RobVar opened on Aug 16, 2026

• Italian language issue

  google-deepmind/gemma#774 · Filter by author ludiromeo opened on Aug 15, 2026

• [Bug] Hackable Diffusion SFT semantics differ from the DiffusionGemma Technical Report

  google-deepmind/gemma#773 · Filter by author DriX-727 opened on Aug 11, 2026

• Missing escape sequence allows to inject a system message

  google-deepmind/gemma#768 · Filter by author pavelai opened on Aug 6, 2026
  1

• Verify evals on Papers with Code

  google-deepmind/gemma#738 · Filter by author NielsRogge opened on Jul 22, 2026

• gemma-4-12B-it: Infinite thinking loop, system prompt non-adherence (>10k tokens), and context loss in multi-turn conversations

  google-deepmind/gemma#727 · Filter by author titansonico10 opened on Jul 14, 2026
  5

• SRT-Sunstone: Cross-modal discourse readout demo on Gemma-4-31B-it

  google-deepmind/gemma#723 · Filter by author space-bacon opened on Jul 9, 2026

• Hallucination/Safety: Model confidently recommends malicious domain (mcped.org)

  google-deepmind/gemma#720 · Filter by author tbjcurtis-eng opened on Jul 8, 2026

• Support Gemma 4 12B Unified in the JAX package

  google-deepmind/gemma#709 · Filter by author lukaemon opened on Jun 27, 2026

• [BUG]: "Wait - I've made a mistake in the logic."

  google-deepmind/gemma#701 · Filter by author asgs opened on Jun 18, 2026
  2

• [Bug]: Context Exhaustion and VRAM Spikes in KV Cache & SamplerLoop

  google-deepmind/gemma#675 · Filter by author prince-shakyaa opened on Jun 7, 2026

• php add

  google-deepmind/gemma#672 · Filter by author raedalmalki3 opened on Jun 5, 2026
  1

• Proposal: Symbolic Activation Framework — Eliminate KV Cache Context Rot

  google-deepmind/gemma#666 · Filter by author gymaira1990-jpg opened on May 28, 2026
  2

• Problem: Commit 5621d2d

  google-deepmind/gemma#662 · Filter by author darknecrocities opened on May 25, 2026

• macOS Apple Silicon: C++ Segfault during test shutdown due to TF GCS client (tokenizer caching)

  google-deepmind/gemma#660 · Filter by author prince-shakyaa opened on May 23, 2026

• Bug: Missing f-prefix in error message in _normalize_token() makes stop_tokens debugging impossible

  google-deepmind/gemma#658 · Filter by author prince-shakyaa opened on May 23, 2026

• Bug: Test suite fails on macOS Apple Silicon & Windows — does not exist

  google-deepmind/gemma#657 · Filter by author prince-shakyaa opened on May 23, 2026

• Gemma 4 E2B emits parenthetical "word(suffix)" continuations under lower-temperature sampling

  google-deepmind/gemma#647 · Filter by author UncreatedLabs opened on May 13, 2026
  2

• Bug: float division instead of integer division in GQA reshape causes silent shape truncation

  google-deepmind/gemma#641 · Filter by author tomohiro86 opened on May 4, 2026

• Gemma 4 appears to have been officially released on 2 April 2026, but the README still lists it as ‘Coming soon’.

  google-deepmind/gemma#639 · Filter by author Pritesh928 opened on Apr 30, 2026
  1

• [Feature request]: Eliminate pre-attention RMSNorm in Gemma 4 via scale invariance + weight folding

  google-deepmind/gemma#638 · Filter by author NilsGraf opened on Apr 29, 2026
  3