Merry commited on
Commit
2d860e7
1 Parent(s): 921b9a8

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +8 -3
README.md CHANGED
@@ -2,11 +2,16 @@
2
  license: apache-2.0
3
  ---
4
 
5
- **NOTE:** Tested with KoboldCpp. I can't promise that it will work with other frontends. Use at your own risk!
 
6
 
7
- **2023-04-20:** Converted using ggerganov/ggml's "stablelm" conversion script and quantization code as of [commit 05f3079](https://github.com/ggerganov/ggml/tree/05f307971862b83df12fada0c42ee027ba5a82b5/examples/stablelm).
8
 
9
- **2023-04-30:** Converted using ggerganov/ggml's "stablelm" conversion script and quantization code as of [commit 5dd92f4](https://github.com/ggerganov/ggml/tree/5dd92f421ee44f18b8fde0afbf5ca8fc7bf93841/examples/stablelm).
 
 
 
 
10
 
11
  These conversions are based off of the updated Pythia Deduped checkpoints, not the original v0 trainings.
12
 
 
2
  license: apache-2.0
3
  ---
4
 
5
+ # **NOTE:** Tested with KoboldCpp.
6
+ ## (I can't promise that it will work with other frontends, if at all. Use at your own risk!)
7
 
8
+ **2023-04-20:** Includes q4_3 versions of all Pythia Deduped models. Converted using ggerganov/ggml's "stablelm" conversion script and quantization code as of [commit 05f3079](https://github.com/ggerganov/ggml/tree/05f307971862b83df12fada0c42ee027ba5a82b5/examples/stablelm).
9
 
10
+ **2023-04-30:** Includes q5_0, q5_1, and q8_0 versions of Pythia Deduped models up to 2.8B. Converted using ggerganov/ggml's "stablelm" conversion script and quantization code as of [commit 5dd92f4](https://github.com/ggerganov/ggml/tree/5dd92f421ee44f18b8fde0afbf5ca8fc7bf93841/examples/stablelm).
11
+
12
+ **2023-05-06:** Includes q4_0 and q4_2 versions of Pythia Deduped models up to 2.8B. Converted using ggerganov/ggml's "stablelm" conversion script and quantization code as of [commit ff6e03c](https://github.com/ggerganov/ggml/tree/ff6e03cbcd9bf6e9fa41d49f2495c042efae4dc6/examples/stablelm).
13
+
14
+ They're separated by date and commit so it's easier to track any breaking changes. The last two go up to 2.8B because it's time-consuming (and unreliable even overnight) for me to upload q4_0, q5_0, q5_1, and q8_0 for the 6.9B and 12B conversions. If you have the hardware for a 7B+ model, I would recommend the currently-training [OpenLLaMA 7B](https://huggingface.co/vihangd/open_llama_7b_300bt_ggml).
15
 
16
  These conversions are based off of the updated Pythia Deduped checkpoints, not the original v0 trainings.
17