Upload weight: checkpoints/training/phase3_all_step1000/trainer_state.pt

Upload weight: checkpoints/training/phase3_all_step1000/lora/adapter_model.safetensors

Upload weight: checkpoints/training/phase3_all_step1000/fusion/fusion.pt

Upload weight: checkpoints/training/crash_checkpoint/vision/projection.pt

Upload weight: checkpoints/training/crash_checkpoint/lora/adapter_model.safetensors

Upload weight: checkpoints/training/crash_checkpoint/fusion/fusion.pt

Upload weight: checkpoints/phase3_all_step1500/vision/projection.pt

Upload weight: checkpoints/phase3_all_step1500/trainer_state.pt

Upload weight: checkpoints/phase3_all_step1500/lora/adapter_model.safetensors

Upload weight: checkpoints/phase3_all_step1500/fusion/fusion.pt

Upload README.md with huggingface_hub

Upload .gitattributes with huggingface_hub

Upload configs

Upload src

Upload checkpoint: vision

Upload checkpoint: phase3_all_step1500

Upload checkpoint: training/phase3_final

Upload checkpoint: training/phase3_all_step500

Upload checkpoint: training/phase3_all_step1500

Upload checkpoint: training/phase3_all_step1000

Upload checkpoint: training/crash_checkpoint

Session 5: Rebuild frontend as Vite+React 3-panel website builder IDE

hf_space: framework-aware system prompt (Next.js / React / Vue / HTML)

frontend: fix StackBlitz launcher (was opening default Next.js starter)

Tier 2: image upload, vision toggle, StackBlitz/CodeSandbox launchers, UI polish

Tier 1: identity anchor + conversation memory + cleaner output

frontend: persistent auth-blocked state + actionable inline message

trainer: skip completed phases / resume mid-phase + update context

Add HF Space backend and frontend with HF token auth

gitignore: exclude data/websight, graphify artifacts, and chat logs

Force line-buffered stdout for real-time log output in docker

Fix gradient checkpointing: enable_input_require_grads for PEFT without torch.compile

Fix OOM: disable torch.compile, reduce batch sizes, double grad_accumulation

Fix hidden_size: 4096 -> 3584 to match Qwen2.5-Coder-7B-Instruct

Add project context doc and WebSight batch uploader

Add WebSight vision data pipeline: download script, image-aware data loader, phase data routing

Fix Phase 2: fusion layer processes text-only via learnable residual gate for gradient flow

Fix: register LLM as nn.Module submodule so optimizer finds LoRA params

Fix extra_special_tokens: list to dict for transformers 4.55

Add requirements-training.txt for MI300X Docker

Remove HSA_OVERRIDE_GFX_VERSION - ROCm 7.0 native MI300X support

Track large tokenizer files with Git LFS for HuggingFace

Add GPU diagnostic script, fix architecture loading with low_cpu_mem_usage and sync

Fix setup_mi300x.sh for Docker container on MI300X droplet

Fix setup_mi300x.sh step 2 + add project context summary

Fix train.py: mem -> memory on line 225

Day 3 COMPLETE: Full model architecture

Day 2 COMPLETE: 1.48M examples processed, 6GB dataset, WebSight done

Day 1 Complete: Tokenizer setup — Qwen2.5-Coder-7B base + 22 MINDI special tokens (vocab 151,685), wrapper class, full format test

feat: initial project scaffold for MINDI 1.5 Vision-Coder