Fix input auto-zoom on mobile devices

Fix mobile browser address bar hiding the input area (100dvh)

Add responsive design media queries for mobile and tablet

Format Live Environments to match user preference

Add Live Environments table and overarching blockquote summary to README

Overhaul README.md with badges, TOC, and deeper architectural explanations

Fix Hugging Face configuration in README

Revert target HF space back to smart-agro-api

Update target HF space to axiom-backend

Upload model weights via LFS

Add GitHub action to automatically sync to Hugging Face Space

Format README with Mermaid diagrams and tabular layouts

Add performance and system metrics to README

Deep dive into PyTorch neural network architecture and tensor math in README

Deep dive README rewrite explaining entire LLM2 ecosystem

Add dataset curriculum and percentages to README

Write comprehensive project README

Hide chat scrollbar for completely sleek full-screen aesthetic

Revert "Refactor UI to professional minimalist aesthetic (ChatGPT/Claude style)"

Refactor UI to professional minimalist aesthetic (ChatGPT/Claude style)

Upgrade UI: Replace native select with custom premium dropdown component

Prevent header from squishing or moving up during chat scroll

Fix chat stretching and jitter during SSE generation

Make chat interface full screen and flexible

Add comprehensive deep-dive README

Prepare for production deployment: Add Dockerfile and env vars

Delete initial README content

Add n_kv_heads to axiom_v1.0.yaml

Add winning architecture settings to axiom_v1.0.yaml

Use official ConfigLoader in FastAPI backend

Fix GPTConfig YAML parsing in FastAPI backend

Fix backend module resolution for axiom_model

Phase 6: Axiom Web Interface (React + FastAPI)

Phase 6: Restructure into Monorepo (axiom_model & axiom_web)

Fix web retriever ddgs warning and add empty results fallback

Phase 4.5 & 4.6: Unified RAG Architecture with Local, Web, and Hybrid backends

Phase 4 v1.0 Polish: Add --debug_retrieval, rename to min_similarity, abstract PromptBuilder

Phase 4 v1.1: Upgrade RAG with explicit prompting, configurable top-k, thresholding, and metadata citations

Phase 4: Implement RAG pipeline with FAISS and SentenceTransformers

Fix chat.py to dynamically load vocab size from checkpoint

Fix model size mismatch when resuming from SFT checkpoints

Fix KeyError new_vocab_size in train_sft.py

Phase 3.1: Advanced SFT pipeline with heuristics, multi-dataset, and early stopping

Add repetition penalty and top-k filtering

Fix SFT target shifting bug

Fix prompt formatting with end tags

Add prompt argument

Fix tokenizer instantiation

Fix Enum parsing in chat.py

Add chat.py for inference