New features:
- Tokens/second throughput estimates based on GPU memory bandwidth
- Model architecture details table (hidden size, heads, GQA ratio, vocab, max context)
- Context length scaling table showing KV cache at different lengths
- Export Report button to generate shareable text reports
- Tips & Guide tab with VRAM rules of thumb and recommended setups
- GPU bandwidth specs for all hardware (Consumer, Apple Silicon, Cloud)
- Best value AND fastest GPU recommendations
- Quantization quality indicators
- More example models (Hermes, Yi)
- Cross-model comparison set (Llama vs Mistral vs Qwen vs Gemma)
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Add enhanced features: colored GPU tables, cloud costs, more examples
Features added:
- Color-coded status indicators (🟢🟡🔴) for GPU fit
- Separate sections for Consumer GPUs, Apple Silicon, Cloud GPUs
- Expanded cloud GPU options with hourly/daily/monthly costs
- Best value cloud recommendation
- GPU Reference tab with all hardware specs
- 12 popular model examples (Llama, Mistral, Qwen, Gemma, Phi, DeepSeek)
- Quick comparison sets for model families
- Improved memory breakdown tables
- Quantization options with fit indicators
- Soft theme for better readability
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Fix Gradio interactivity: use Tabs wrapper with TabItem pattern
Key fixes:
- Wrap tabs in gr.Tabs() container (required for HF Spaces)
- Use gr.TabItem() instead of gr.Tab() for better compatibility
- Use explicit keyword args (minimum=, maximum=, etc.)
- Add try/except error handling to all functions
- Remove unused pandas dependency
- Use string concatenation instead of f-strings for Python 3.9
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Fix Python 3.9 compatibility and HuggingFace Spaces issues
Critical fixes:
- Add 'from __future__ import annotations' for Python 3.9 compatibility
- Remove type hints that caused SyntaxError on older Python
- Use gr.Tabs() with gr.TabItem() for reliable tab switching
- Remove problematic 'filter' parameter from list_models
- Simplify error handling to return messages instead of raising
- Ensure all Examples components work correctly
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Fix UI bugs: tabs, accordion, examples, and sliders
New features:
- LoRA and QLoRA fine-tuning memory estimation modes
- Model comparison tab for side-by-side VRAM analysis
- Model search with HuggingFace API integration
- Throughput estimation (tokens/sec) per GPU
- Cloud cost estimates (hourly/daily/monthly)
- Flash Attention toggle with memory savings display
- Export results to JSON or plain text
- Tabbed interface (Calculator, Compare, Export)
Updated GPU specs with TFLOPs and hourly costs for cloud instances.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Add training mode, multi-GPU support, expanded GPU database, quantization breakdown, and visual memory chart
