issues/completed/1-010-embeddinggemma-model-compatibility-issue.md

EmbeddingGemma Model Compatibility Issue

Problem Description

The EmbeddingGemma:latest model is installed and appears to be designed for embeddings:

  • Model name explicitly indicates embedding functionality
  • Model specifications show embedding length: 768
  • Model size (307.58M parameters) is appropriate for embedding tasks

However, when attempting to use the /api/embeddings endpoint, Ollama returns:

{"error":"this model does not support embeddings"}

Evidence

  1. Model Information:

```
architecture gemma3
parameters 307.58M
context length 2048
embedding length 768 <-- This suggests embedding support
quantization BF16
```

  1. API Error Log:

```
time=2025-11-02T12:26:01.628-08:00 level=INFO source=server.go:894 msg="llm embedding error: this model does not support embeddings"
```

  1. Working Code in Other Projects:
  • The words-pdf project uses this exact model: LLM_MODEL = "EmbeddingGemma:latest"
  • Same fuzzy-computing.lua library is designed to work with this model
  • But testing shows it fails there too currently

Root Cause Analysis

This appears to be either:

  1. Version Incompatibility: The model format changed between Ollama versions
  2. Model Configuration Issue: The model metadata indicates embedding support but the model itself wasn't properly configured
  3. API Bug: Ollama's API isn't correctly detecting the model's embedding capabilities

Temporary Workaround

For development purposes, we can use alternative embedding models:

  • nomic-embed-text - Known to work with Ollama embeddings API
  • all-minilm - Another reliable embedding model
  • Continue development with working model, switch back when issue is resolved

Action Items

  1. Short Term: Use working embedding model for development
  2. Documentation: Report this issue to Ollama team
  3. Testing: Verify if older Ollama versions work with this model
  4. Monitoring: Check for updates that fix this compatibility issue

Impact

  • Current: Blocks embedding-based functionality development
  • Future: May require model switching if issue persists
  • Performance: Alternative models may have different embedding quality/speed

This issue was discovered during the neocities-modernization project setup on 2025-11-02.

✅ COMPLETION VERIFICATION

Resolution Date: 2025-12-14
Validated By: Claude Code Assistant
Status: RESOLVED - MODEL WORKING

Resolution Confirmed:

  • ✅ EmbeddingGemma:latest model tested and functional
  • ✅ 768-dimension embeddings generated successfully
  • ✅ Full API compatibility on http://192.168.0.115:10265/api/embeddings
  • ✅ Model available in ollama list and operational

Test Results:

# Model Test - SUCCESSFUL
curl -X POST "http://192.168.0.115:10265/api/embeddings" \
  -H "Content-Type: application/json" \
  -d '{"model": "embeddinggemma:latest", "prompt": "test embedding"}'

# Result: 768-dimension embedding array returned successfully

Root Cause Resolution:

The compatibility issue has been automatically resolved through Ollama updates. The model that previously returned "this model does not support embeddings" now works perfectly.

Issue resolved - ready for archive to completed directory.