llm-forge-new 0.1.0__tar.gz

llm_forge_new-0.1.0/.claude/hooks/safety-check.sh

@@ -0,0 +1,35 @@
+#!/bin/bash
+# Safety check hook — blocks dangerous operations in Claude Code
+
+INPUT="$TOOL_INPUT_command"
+
+# Block force pushes to main (use --force-with-lease on feature branches only)
+if echo "$INPUT" | grep -q "git push.*--force"; then
+  echo "BLOCKED: Force push detected."
+  echo "Use --force-with-lease on feature branches only."
+  echo "Never force push to main."
+  exit 1
+fi
+
+# Block recursive deletion of critical directories
+if echo "$INPUT" | grep -qE "rm.*-rf.*(src|tests|configs|\.claude)"; then
+  echo "BLOCKED: Recursive deletion of critical directory detected."
+  echo "Directories src/, tests/, configs/, and .claude/ are protected."
+  exit 1
+fi
+
+# Block accidental deletion of the database
+if echo "$INPUT" | grep -qE "rm.*metrics\.db"; then
+  echo "BLOCKED: Deletion of metrics database detected."
+  echo "The metrics database contains historical training data."
+  exit 1
+fi
+
+# Warn about pip install without --break-system-packages
+if echo "$INPUT" | grep -q "pip install" && ! echo "$INPUT" | grep -q "break-system-packages"; then
+  echo "WARNING: Consider using --break-system-packages flag with pip install."
+  echo "Continuing anyway..."
+fi
+
+# All checks passed
+exit 0

llm_forge_new-0.1.0/.claude/settings.json

@@ -0,0 +1,26 @@
+{
+  "hooks": {
+    "PostToolUse": [
+      {
+        "matcher": "Edit|Write",
+        "hooks": [
+          {
+            "type": "command",
+            "command": "cd \"$CLAUDE_PROJECT_DIR\" && python -m ruff check --fix $TOOL_INPUT_path 2>/dev/null; exit 0"
+          }
+        ]
+      }
+    ],
+    "PreToolUse": [
+      {
+        "matcher": "Bash",
+        "hooks": [
+          {
+            "type": "command",
+            "command": "bash .claude/hooks/safety-check.sh"
+          }
+        ]
+      }
+    ]
+  }
+}

llm_forge_new-0.1.0/.github/ISSUE_TEMPLATE/bug_report.yml

@@ -0,0 +1,54 @@
+name: Bug Report
+description: Report a bug in llm-forge
+title: "[Bug]: "
+labels: ["bug"]
+body:
+  - type: markdown
+    attributes:
+      value: |
+        Thanks for reporting a bug! Please fill out the information below.
+  - type: textarea
+    id: description
+    attributes:
+      label: Description
+      description: A clear description of the bug.
+    validations:
+      required: true
+  - type: textarea
+    id: reproduce
+    attributes:
+      label: Steps to Reproduce
+      description: Steps to reproduce the behavior.
+      placeholder: |
+        1. Create config with...
+        2. Run `llm-forge train --config ...`
+        3. See error...
+    validations:
+      required: true
+  - type: textarea
+    id: expected
+    attributes:
+      label: Expected Behavior
+      description: What you expected to happen.
+    validations:
+      required: true
+  - type: textarea
+    id: config
+    attributes:
+      label: Configuration (YAML)
+      description: Your YAML config file (remove any sensitive info).
+      render: yaml
+  - type: textarea
+    id: environment
+    attributes:
+      label: Environment
+      description: Output of `llm-forge info`
+      render: text
+    validations:
+      required: true
+  - type: textarea
+    id: logs
+    attributes:
+      label: Error Logs
+      description: Relevant error messages or stack traces.
+      render: text

llm_forge_new-0.1.0/.github/ISSUE_TEMPLATE/feature_request.yml

@@ -0,0 +1,42 @@
+name: Feature Request
+description: Suggest a new feature for llm-forge
+title: "[Feature]: "
+labels: ["enhancement"]
+body:
+  - type: markdown
+    attributes:
+      value: |
+        Thanks for suggesting a feature! Please describe what you'd like.
+  - type: textarea
+    id: description
+    attributes:
+      label: Feature Description
+      description: A clear description of the feature you'd like.
+    validations:
+      required: true
+  - type: textarea
+    id: motivation
+    attributes:
+      label: Motivation
+      description: Why would this feature be useful?
+    validations:
+      required: true
+  - type: textarea
+    id: alternatives
+    attributes:
+      label: Alternatives Considered
+      description: Any alternative solutions you've considered.
+  - type: dropdown
+    id: area
+    attributes:
+      label: Component
+      options:
+        - Configuration
+        - Data Pipeline
+        - Training
+        - Evaluation
+        - RAG
+        - Serving
+        - CLI
+        - Documentation
+        - Other

llm_forge_new-0.1.0/.github/PULL_REQUEST_TEMPLATE.md

@@ -0,0 +1,20 @@
+## Summary
+
+<!-- Brief description of changes -->
+
+## Changes
+
+-
+
+## Testing
+
+- [ ] Added/updated tests
+- [ ] All tests pass (`make test`)
+- [ ] Linting passes (`make lint`)
+
+## Type of Change
+
+- [ ] Bug fix
+- [ ] New feature
+- [ ] Breaking change
+- [ ] Documentation update

llm_forge_new-0.1.0/.github/workflows/ci.yml

@@ -0,0 +1,73 @@
+name: CI
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+    branches: [main]
+
+permissions:
+  contents: read
+
+env:
+  FORCE_JAVASCRIPT_ACTIONS_TO_NODE24: true
+
+jobs:
+  lint:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - name: Set up Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: "3.11"
+      - name: Install dependencies
+        run: |
+          pip install ruff mypy
+      - name: Run ruff check
+        run: ruff check src/ tests/
+      - name: Run ruff format check
+        run: ruff format --check src/ tests/
+
+  test-cpu:
+    runs-on: ubuntu-latest
+    strategy:
+      matrix:
+        python-version: ["3.10", "3.11", "3.12"]
+    steps:
+      - uses: actions/checkout@v4
+      - name: Set up Python ${{ matrix.python-version }}
+        uses: actions/setup-python@v5
+        with:
+          python-version: ${{ matrix.python-version }}
+      - name: Install dependencies
+        run: |
+          pip install -e ".[dev]"
+      - name: Run tests
+        env:
+          NO_COLOR: "1"
+        run: |
+          pytest tests/ -v --cov=llm_forge --cov-report=xml \
+            --ignore=tests/test_training/test_gpu.py \
+            -m "not gpu and not slow"
+      - name: Upload coverage
+        if: matrix.python-version == '3.11'
+        uses: codecov/codecov-action@v5
+        with:
+          files: ./coverage.xml
+
+  build:
+    runs-on: ubuntu-latest
+    needs: [lint, test-cpu]
+    steps:
+      - uses: actions/checkout@v4
+      - name: Set up Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: "3.11"
+      - name: Install build tools
+        run: pip install build twine
+      - name: Build package
+        run: python -m build
+      - name: Check package
+        run: twine check dist/*

llm_forge_new-0.1.0/.github/workflows/gpu-tests.yml

@@ -0,0 +1,31 @@
+name: GPU Tests
+
+on:
+  pull_request:
+    types: [labeled]
+
+permissions:
+  contents: read
+
+env:
+  FORCE_JAVASCRIPT_ACTIONS_TO_NODE24: true
+
+jobs:
+  test-gpu:
+    if: github.event.label.name == 'gpu-test'
+    runs-on: self-hosted
+    steps:
+      - uses: actions/checkout@v4
+      - name: Set up Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: "3.11"
+      - name: Install dependencies
+        run: |
+          pip install -e ".[all]"
+      - name: Verify GPU
+        run: |
+          python -c "import torch; print(f'CUDA: {torch.cuda.is_available()}, GPUs: {torch.cuda.device_count()}')"
+      - name: Run GPU tests
+        run: |
+          pytest tests/ -v -m gpu --timeout=600

llm_forge_new-0.1.0/.github/workflows/release.yml

@@ -0,0 +1,58 @@
+name: Release
+
+on:
+  push:
+    tags:
+      - "v*"
+
+permissions: {}
+
+env:
+  FORCE_JAVASCRIPT_ACTIONS_TO_NODE24: true
+
+jobs:
+  publish:
+    runs-on: ubuntu-latest
+    environment: release
+    permissions:
+      id-token: write
+      contents: read
+    steps:
+      - uses: actions/checkout@v4
+      - name: Set up Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: "3.11"
+      - name: Install build tools
+        run: pip install build
+      - name: Build package
+        run: python -m build
+      - name: Publish to PyPI
+        uses: pypa/gh-action-pypi-publish@release/v1
+
+  github-release:
+    runs-on: ubuntu-latest
+    needs: publish
+    permissions:
+      contents: write
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
+      - name: Generate changelog
+        id: changelog
+        run: |
+          PREV_TAG=$(git tag --sort=-version:refname | sed -n '2p')
+          if [ -z "$PREV_TAG" ]; then
+            echo "body=Initial release" >> $GITHUB_OUTPUT
+          else
+            CHANGES=$(git log --pretty=format:"- %s" ${PREV_TAG}..HEAD)
+            echo "body<<EOF" >> $GITHUB_OUTPUT
+            echo "$CHANGES" >> $GITHUB_OUTPUT
+            echo "EOF" >> $GITHUB_OUTPUT
+          fi
+      - name: Create GitHub Release
+        uses: softprops/action-gh-release@v2
+        with:
+          body: ${{ steps.changelog.outputs.body }}
+          generate_release_notes: true

llm_forge_new-0.1.0/.gitignore

@@ -0,0 +1,88 @@
+# Python
+__pycache__/
+*.py[cod]
+*$py.class
+*.so
+*.egg-info/
+*.egg
+dist/
+build/
+.eggs/
+
+# Virtual environments
+.venv/
+venv/
+env/
+
+# IDE
+.vscode/
+.idea/
+*.swp
+*.swo
+*~
+
+# OS
+.DS_Store
+Thumbs.db
+
+# Project
+output/
+outputs/
+checkpoints/
+logs/
+wandb/
+*.ckpt
+*.pt
+*.bin
+*.safetensors
+*.gguf
+
+# Data (user training data — not committed)
+/data/
+*.jsonl
+*.parquet
+*.csv
+# Exception: example data IS committed
+!examples/data/*.jsonl
+!examples/data/*.txt
+
+# Environment and secrets
+.env
+.env.*
+*.key
+*.pem
+*.p12
+*.secret
+credentials.json
+service-account*.json
+
+# HuggingFace token cache
+.cache/huggingface/token
+
+# Eval results (large, machine-specific)
+eval_results/
+
+# Models (large files)
+models/
+*.model
+lid.176.bin
+
+# ChromaDB
+chroma_db/
+
+# Coverage
+.coverage
+htmlcov/
+coverage.xml
+
+# Jupyter
+.ipynb_checkpoints/
+
+# Benchmark results (machine-specific paths)
+hopper_benchmark/
+
+# Metrics database
+metrics.db
+
+# Claude project memory (personal context)
+.claude/projects/

llm_forge_new-0.1.0/CHANGELOG.md

@@ -0,0 +1,54 @@
+# Changelog
+
+All notable changes to llm-forge will be documented in this file.
+
+The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
+and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
+
+## [0.1.0] - 2026-03-19
+
+### Added
+- Initial public release of llm-forge
+- YAML-first configuration system with Pydantic v2 validation (22 config classes, 183 fields)
+- Universal data loader (JSONL, CSV, Parquet, PDF, DOCX, HTML, HuggingFace datasets)
+- Full data cleaning pipeline (unicode fixing, language filtering, heuristic filters, quality classification, toxicity filtering, PII redaction, 3-tier deduplication)
+- LoRA, QLoRA, and full fine-tuning with TRL SFTTrainer
+- Pre-training from scratch with configurable Llama architecture
+- Alignment training: DPO, ORPO, GRPO, and PPO-based RLHF
+- Inference-Time Intervention (ITI) for anti-hallucination with probing and weight baking
+- IFD (Instruction-Following Difficulty) scoring for data quality
+- Refusal augmentation (R-Tuning) for safety training
+- Model merging: linear interpolation, SLERP, and TIES-Merging
+- Synthetic data generation with quality scoring
+- Distributed training orchestration (FSDP, DeepSpeed ZeRO 0-3, Megatron-Core 3D parallelism)
+- FP8 training support for Hopper GPUs via Transformer Engine
+- Hardware auto-detection and config optimization for RTX 3090/4090, A100, H100, Apple Silicon
+- lm-evaluation-harness benchmark integration (MMLU, GSM8K, ARC, HellaSwag, IFEval, etc.)
+- LLM-as-Judge evaluation with configurable criteria
+- Domain evaluation with custom datasets
+- Knowledge retention probes (100 questions across 10 domains)
+- RAG pipeline with hybrid retrieval (dense + BM25), cross-encoder reranking, ChromaDB/FAISS
+- Gradio web dashboard with model browser, training UI, and chat interface
+- FastAPI server with OpenAI-compatible API endpoints
+- vLLM high-throughput serving integration
+- Model export: safetensors, GGUF (16 quantization types), ONNX, AWQ
+- Ollama Modelfile generation with multi-turn chat template support
+- Typer CLI with 13 commands and Rich formatting
+- Interactive training wizard for guided setup
+- Apple Silicon support (MPS + MLX trainer)
+- Mac-specific monitoring (thermal, battery, memory pressure)
+- SLURM job generation for HPC clusters
+- Training callbacks: Rich progress, GPU monitoring, W&B, early stopping, checkpointing
+- Comprehensive error recovery with context-aware suggestions
+- Security module: safetensors validation, credential masking
+- 894 passing tests, 0 failures
+- 20 ready-to-use YAML configs (quickstart, benchmarks, domain examples, cloud deployment)
+- Example training data in Alpaca, ShareGPT, and completion formats
+- GitHub Actions CI/CD (lint, test, build, PyPI release)
+
+### Finance Specialist Model (v7)
+- Trained and deployed to HuggingFace: Venkat9990/finance-specialist-v7
+- Zero catastrophic forgetting: all benchmarks within 0-2% of base model
+- Knowledge-preserving LoRA (r=8, attention-only, LR 1e-5)
+- 97.7% token masking accuracy
+- Available as safetensors (2.4 GB) and GGUF Q4_K_M (763 MB)