richardyoung commited on
Commit
9121eeb
·
verified ·
1 Parent(s): 5db7af7

Upload reproduce/README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. reproduce/README.md +69 -0
reproduce/README.md ADDED
@@ -0,0 +1,69 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Reproduction guide
2
+
3
+ This directory contains the necessary information and assets to reproduce the results obtained during this Heretic run.
4
+
5
+ ## Models
6
+
7
+ - **Base model:** [squ11z1/Mythos-nano](https://huggingface.co/squ11z1/Mythos-nano) (Commit: [`64667ba`](https://huggingface.co/squ11z1/Mythos-nano/commit/64667ba4c7f320348f0398916304d0c22f37c56d))
8
+
9
+ ## Datasets
10
+
11
+ - **Good prompts:** [mlabonne/harmless_alpaca](https://huggingface.co/datasets/mlabonne/harmless_alpaca) (Commit: [`02c6a92`](https://huggingface.co/datasets/mlabonne/harmless_alpaca/commit/02c6a92cfcf11bb0c387334f8146d149d65b587f))
12
+ - **Bad prompts:** [mlabonne/harmful_behaviors](https://huggingface.co/datasets/mlabonne/harmful_behaviors) (Commit: [`01cead0`](https://huggingface.co/datasets/mlabonne/harmful_behaviors/commit/01cead01398926d81f7c52bdb790ee8cf77ebba7))
13
+ - **Good evaluation prompts:** [mlabonne/harmless_alpaca](https://huggingface.co/datasets/mlabonne/harmless_alpaca) (Commit: [`02c6a92`](https://huggingface.co/datasets/mlabonne/harmless_alpaca/commit/02c6a92cfcf11bb0c387334f8146d149d65b587f))
14
+ - **Bad evaluation prompts:** [mlabonne/harmful_behaviors](https://huggingface.co/datasets/mlabonne/harmful_behaviors) (Commit: [`01cead0`](https://huggingface.co/datasets/mlabonne/harmful_behaviors/commit/01cead01398926d81f7c52bdb790ee8cf77ebba7))
15
+
16
+ ## Selected trial
17
+
18
+ - **Trial number:** 188
19
+ - **KL divergence:** 0.060961
20
+ - **Refusals:** 13/100
21
+
22
+ ## System
23
+
24
+ - **Python:** 3.12.3 (CPython, GCC 13.3.0) [Virtualenv/Venv]
25
+ - **Operating system:** Linux-6.8.0-124-generic-x86_64-with-glibc2.39 (x86_64)
26
+ - **CPU:** AMD Ryzen 9 7950X 16-Core Processor
27
+
28
+ ### Accelerators
29
+
30
+ - **CUDA:** Detected 1 device(s) (23.51 GB total VRAM)
31
+ - **CUDA Version:** 13.0
32
+ - **Driver Version:** 580.159.03
33
+ - **Devices:**
34
+ - **CUDA 0:** NVIDIA GeForce RTX 4090 (23.51 GB)
35
+
36
+ ## Environment
37
+
38
+ - **Heretic:** v1.4.0 (Origin: PyPI)
39
+ - **PyTorch:** 2.12.1+cu130
40
+ - **Other dependencies:** See [`requirements.txt`](requirements.txt).
41
+
42
+ ## Contents of this directory
43
+
44
+ - [`requirements.txt`](requirements.txt): The exact versions of all Python packages.
45
+ - [`config.toml`](config.toml): The exact configuration used, including the RNG seed.
46
+ - [`squ11z1--Mythos-nano.jsonl`](squ11z1--Mythos-nano.jsonl): The Optuna study journal containing the history of all trials.
47
+ - [`SHA256SUMS`](SHA256SUMS): Cryptographic hashes for all weight files.
48
+ - [`reproduce.json`](reproduce.json): A machine-readable file containing all reproducibility information.
49
+
50
+ ## How to reproduce
51
+
52
+ > [!TIP]
53
+ > You can automate this process, including all verification steps, by downloading the `reproduce.json` file and running
54
+ > `heretic --reproduce reproduce.json`.
55
+
56
+ 1. Ensure your system matches the specifications in the **System** section above. Exact reproducibility is only guaranteed if all aspects of your system are identical to the one the model was originally generated on.
57
+ 1. Install the exact version of Heretic indicated in the **Environment** section above, from its original source.
58
+ 1. Install the packages listed in `requirements.txt`: `pip install -r requirements.txt`
59
+ 1. Install the correct version of PyTorch: `pip install torch==2.12.1+cu130 --index-url https://download.pytorch.org/whl/cu130`
60
+ 1. Place the provided `config.toml` in your working directory.
61
+ 1. Run Heretic without any additional arguments: `heretic`
62
+ 1. Wait for the run to finish, then select trial **188** and export the model.
63
+ 1. Verify that the weight files have been exactly reproduced by comparing their SHA-256 hashes against those in `SHA256SUMS`:
64
+ `sha256sum -c SHA256SUMS` (or look at the hashes online if you uploaded to Hugging Face)
65
+
66
+ > [!TIP]
67
+ > To use the included Optuna study journal `squ11z1--Mythos-nano.jsonl`, place it in the checkpoints directory (usually `checkpoints/`) before running Heretic.
68
+ >
69
+ > This allows you to export other models from the Pareto front, or to run additional trials without having to re-run the stored trials.