Add gguf_file parameter to tokenizer loading and introduce diagnostic script for GGUF validation
ndc8
6f81ff7
Fix GGUF filename in environment variable and update comment in requirements
ndc8
a2a4e98
Add protobuf requirement for GGUF model loading
ndc8
0c37db0
Refactor application to implement GGUF backend with native transformers support; update requirements and add GGUF-specific entry point
ndc8
6e96e6e
Refactor model loading to utilize accelerate for device management; add test script to verify loading fix and prevent device conflicts
ndc8
8a3c5dd
Refactor application to use lightweight backend; update requirements and add memory analysis script for optimized model configuration
ndc8
a4ee3a6
Update Dockerfile and application entry point for GGUF backend; optimize memory usage in model parameters and requirements
ndc8
358e717
d
ndc8
994c0b4
Refactor backend service to support Gemma 3n model and update requirements; remove obsolete test script and add new dependency tests
ndc8
4b4e9ed
c
ndc8
4f67c26
b
ndc8
557b035
l
ndc8
819080c
a
ndc8
6496777
Add metadata section to README for project details
ndc8
ae13708
aa
ndc8
65edee9
Add scripts for converting and generating UltraChat-style SFT datasetβ
ndc8
7ecd130
- Implement `convert_ultrachat_to_text.py` to convert UltraChat records into a single text format for training.
- Implement `generate_ultrachat_sft.py` to generate synthetic UltraChat-style SFT datasets with customizable parameters.
update
ndc8
91181f3
Update Dockerfile: Add missing system dependencies for build
ndc8
9fe463f
rabbit-ed
ndc8
3960f0f
Cleanup: Remove unnecessary files and update .gitignore
ndc8
78b611a
update
ndc8
4ecf54e
chg model
ndc8
375ade4
change to adapter
ndc8
0c9134e
try
ndc8
db8cd85
upd
ndc8
cb5d5f8
update to use unsloth + mistral
ndc8
172b424
change model
ndc8
8208c22
try #1
ndc8
1ba257c
upd
ndc8
8d9c495
update
ndc8
3239c69
Fix model dependencies and warningsβ
ndc8
97bafdb
- Add protobuf>=3.20.0 to requirements.txt (required for DeepSeek tokenizer)
- Add warning suppression for FutureWarning and rope_scaling messages
- Suppress slow image processor warnings
- These changes address the tokenizer loading issues with the new DeepSeek model
Update model to unsloth/DeepSeek-R1-0528-Qwen3-8B-GGUFβ
ndc8
c6f7b75
- Changed current_model in backend_service.py
- Updated ChatCompletionRequest default model
- Updated all test files to use new model
- Updated requirements.txt comment
- DeepSeek R1-0528 is a more modern and capable model than the previous gemma model
Revert model back to gemma-3n-E4B-it-GGUFβ
ndc8
88c3c6c
- Keep the original model as requested
- Maintain the fix for deprecated use_auth_token parameter
Fix: Update to valid HuggingFace model and fix deprecation warningsβ
ndc8
04d695c
- Changed model from 'gemma-3n-E4B-it-GGUF' to 'microsoft/DialoGPT-medium'
- Fixed deprecated 'use_auth_token' parameter to 'token'
- Updated test file to use the correct model name
upd
ndc8
2cd680b
fix
ndc8
255b6fc
fix
ndc8
84eb396
update
ndc8
e46cec3
upd
ndc8
68f41f4
update
ndc8
83df634
fix
ndc8
4599528
Ignore Hugging Face cache directory
ndc8
385d87b
fix
ndc8
e6598e6
Add API response endpoint and tests
ndc8
09c9042
Set gemma-3n-E4B-it-GGUF as main model for all text generation endpoints
- Update README metadata to use 'docker' SDK (required by HF Spaces)
- Add Dockerfile for containerized deployment
- Set correct app_port for FastAPI service
- Remove unsupported fastapi SDK configuration
This enables deployment to HuggingFace Spaces with Docker runtime.
π Add multimodal AI capabilities with image-text-to-text pipelineβ
ndc8
4e10023
β¨ Features:
- Integrated transformers pipeline for image analysis
- Added Salesforce BLIP model for image captioning
- Enhanced FastAPI backend with multimodal support
- OpenAI Vision API compatible message format
- Dual model architecture (text + vision)
- Comprehensive testing suite
π§ Technical:
- Added Pydantic models for multimodal content
- Image processing utilities for URL handling
- Automatic request routing (text-only vs multimodal)
- Error handling and fallback mechanisms
- Updated dependencies for transformers, torch, PIL
π Documentation:
- Complete integration guide
- Updated README with multimodal examples
- Comprehensive testing documentation
- Usage examples for all endpoints
β Tested and validated:
- All 4 test categories passing
- Text-only chat functionality preserved
- Image analysis working perfectly
- Multimodal chat combining image + text