attacklm 0.8.2__tar.gz → 0.8.3__tar.gz

{attacklm-0.8.2 → attacklm-0.8.3}/CHANGELOG.md

@@ -1,7 +1,24 @@
 # Changelog
 
+## [0.8.3] — 2026-07-01 — Training pair evolution
+
+### Added
+- **Training pair evolution**: New `scripts/evolve_pairs.py` with 3 strategies to expand short training pairs into longer, richer examples
+  - **Evol-Instruct**: Rewrites answers with deeper reasoning, edge cases, and detection artifacts (3-5x longer)
+  - **Multi-turn Decomposition**: Breaks Q&A into 3-5 turn conversations for better training flow
+  - **Chain-of-Thought Injection**: Adds explicit reasoning steps before final answers
+- **Quality filtering**: `scripts/filter_evolved.py` validates evolved pairs (structure, length, provenance, dedup)
+- **`--evolved-ratio` flag**: Mix evolved pairs into training at configurable ratio in `train_all.py` and `train_template.py`
+- **`scripts/evolved_mixer.py`**: Standalone mixer module for evolved pair integration
+
+### Changed
+- **Agent models upgraded**: glm-5.1 → glm-5.2, minimax-m2.7 → minimax-m3, added kimi-k2.7-code
+
+---
+
 ## [0.8.1] — 2026-06-30 — Zero-config dataset init
 
+
 ### Added
 - **Zero-config init**: `attacklm init` now downloads a pre-built dataset tarball (~50 MB) from GitHub releases by default. No git clone, no extractors, no manual setup required.
 - **`--from-source` flag**: `attacklm init --from-source` preserves the old clone+extract pipeline for developers who want to rebuild from upstream repos.

{attacklm-0.8.2 → attacklm-0.8.3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: attacklm
-Version: 0.8.2
+Version: 0.8.3
 Summary: QLoRA-fine-tuned Qwen2.5-Coder for offensive/defensive security and AI red-teaming
 Project-URL: Homepage, https://github.com/Veedubin/AttackLM
 Project-URL: Repository, https://github.com/Veedubin/AttackLM
@@ -287,6 +287,14 @@ By default, AttackLM leverages PyTorch's built-in `torch.backends.cuda.enable_me
 
 - **Comprehensive Security Corpus**: 24,652 high-quality training pairs across 18 distinct security sources.
 - **Advanced Training Methods**: Support for QLoRA, GaLore, Q-GaLore, Spectrum, and PiSSA to enable training of large models on consumer hardware.
+- **Training Pair Evolution**: New capability to synthetically expand short, factual pairs into complex reasoning examples using three specialized strategies:
+
+| Strategy | Approach | Impact |
+| :--- | :--- | :--- |
+| **Evol-Instruct** | Rewrites responses with deeper reasoning and edge cases | 3-5x increase in response length/depth |
+| **Multi-turn** | Decomposes Q&A into interactive conversations | Improved conversational flow and context |
+| **CoT Injection** | Adds explicit "Chain-of-Thought" reasoning steps | Higher logical consistency in complex tasks |
+
 - **Zero-Config Setup**: One-shot `init` command that handles dataset retrieval, extraction, and bucket organization.
 - **Anti-Bias Balancing**: Integrated balancing engine to ensure the model learns diverse tactics rather than just the most voluminous sources.
 - **Provenance Tracking**: Strict per-source attribution and license tracking for every record in the dataset.
@@ -314,11 +322,20 @@ attacklm balance --profile 7b-16gb --preset red-team
 ### Model Training
 `attacklm train`
 The core training engine. Supports Qwen2.5-Coder 3B and 7B base models.
+
+**Standard Training**
 ```bash
 # Train a single model on the entire balanced dataset
 attacklm train -- --dataset all --epochs 10 --lora-r 16 --use-galore
 ```
 
+**Training with Evolved Pairs**
+Use the `--evolved-ratio` flag to mix synthetically evolved high-reasoning pairs into your training set.
+```bash
+# Train with 20% evolved pairs for better reasoning depth
+attacklm train -- --dataset all --evolved-ratio 0.2 --epochs 10
+```
+
 ### Deployment & Testing
 `attacklm build`
 Merge LoRA adapters and convert to GGUF for local deployment.
@@ -384,6 +401,7 @@ This hierarchy ensures that the pipeline can be rebuilt from upstream sources wi
 | :--- | :--- |
 | `attacklm train` | Train a model (QLoRA, GaLore, Q-GaLore, Spectrum, PiSSA) |
 | `attacklm train --dataset all` | Train all buckets combined |
+| `attacklm train --evolved-ratio 0.2` | Mix evolved reasoning pairs into training |
 | `attacklm train --hpo` | Run Hyper-Parameter Optimization sweep |
 | `attacklm init` | Initialize dataset: download pre-built or clone $\rightarrow$ extract $\rightarrow$ attribute |
 | `attacklm balance` | Build a balanced subset of buckets to prevent overfitting |

{attacklm-0.8.2 → attacklm-0.8.3}/README.md

@@ -89,6 +89,14 @@ By default, AttackLM leverages PyTorch's built-in `torch.backends.cuda.enable_me
 
 - **Comprehensive Security Corpus**: 24,652 high-quality training pairs across 18 distinct security sources.
 - **Advanced Training Methods**: Support for QLoRA, GaLore, Q-GaLore, Spectrum, and PiSSA to enable training of large models on consumer hardware.
+- **Training Pair Evolution**: New capability to synthetically expand short, factual pairs into complex reasoning examples using three specialized strategies:
+
+| Strategy | Approach | Impact |
+| :--- | :--- | :--- |
+| **Evol-Instruct** | Rewrites responses with deeper reasoning and edge cases | 3-5x increase in response length/depth |
+| **Multi-turn** | Decomposes Q&A into interactive conversations | Improved conversational flow and context |
+| **CoT Injection** | Adds explicit "Chain-of-Thought" reasoning steps | Higher logical consistency in complex tasks |
+
 - **Zero-Config Setup**: One-shot `init` command that handles dataset retrieval, extraction, and bucket organization.
 - **Anti-Bias Balancing**: Integrated balancing engine to ensure the model learns diverse tactics rather than just the most voluminous sources.
 - **Provenance Tracking**: Strict per-source attribution and license tracking for every record in the dataset.
@@ -116,11 +124,20 @@ attacklm balance --profile 7b-16gb --preset red-team
 ### Model Training
 `attacklm train`
 The core training engine. Supports Qwen2.5-Coder 3B and 7B base models.
+
+**Standard Training**
 ```bash
 # Train a single model on the entire balanced dataset
 attacklm train -- --dataset all --epochs 10 --lora-r 16 --use-galore
 ```
 
+**Training with Evolved Pairs**
+Use the `--evolved-ratio` flag to mix synthetically evolved high-reasoning pairs into your training set.
+```bash
+# Train with 20% evolved pairs for better reasoning depth
+attacklm train -- --dataset all --evolved-ratio 0.2 --epochs 10
+```
+
 ### Deployment & Testing
 `attacklm build`
 Merge LoRA adapters and convert to GGUF for local deployment.
@@ -186,6 +203,7 @@ This hierarchy ensures that the pipeline can be rebuilt from upstream sources wi
 | :--- | :--- |
 | `attacklm train` | Train a model (QLoRA, GaLore, Q-GaLore, Spectrum, PiSSA) |
 | `attacklm train --dataset all` | Train all buckets combined |
+| `attacklm train --evolved-ratio 0.2` | Mix evolved reasoning pairs into training |
 | `attacklm train --hpo` | Run Hyper-Parameter Optimization sweep |
 | `attacklm init` | Initialize dataset: download pre-built or clone $\rightarrow$ extract $\rightarrow$ attribute |
 | `attacklm balance` | Build a balanced subset of buckets to prevent overfitting |