Text Generation
Transformers
PyTorch
mistral
conversational
text-generation-inference
Inference Endpoints
LoneStriker commited on
Commit
df3a89f
1 Parent(s): 70aec39

Upload folder using huggingface_hub

Browse files
Files changed (2) hide show
  1. README.md +32 -0
  2. output.safetensors +1 -1
README.md CHANGED
@@ -5,6 +5,34 @@ datasets:
5
  pipeline_tag: text-generation
6
  ---
7
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8
  ### Dataset:
9
  Training dataset: [snorkelai/Snorkel-Mistral-PairRM-DPO-Dataset](https://huggingface.co/datasets/snorkelai/Snorkel-Mistral-PairRM-DPO-Dataset)
10
 
@@ -19,6 +47,10 @@ We utilize ONLY the prompts from [UltraFeedback](https://huggingface.co/datasets
19
  This overview provides a high-level summary of our approach.
20
  We plan to release more detailed results and findings in the coming weeks on the [Snorkel blog.](https://snorkel.ai/blog/)
21
 
 
 
 
 
22
  ### Training recipe:
23
  - The provided data is formatted to be compatible with the Hugging Face's [Zephyr recipe](https://github.com/huggingface/alignment-handbook/tree/main/recipes/zephyr-7b-beta).
24
  We executed the n_th DPO iteration using the "train/test_iteration_{n}".
 
5
  pipeline_tag: text-generation
6
  ---
7
 
8
+ We offer a temporary HF space for everyone to try out the model: -> [**Snorkel-Mistral-PairRM-DPO Space**](https://huggingface.co/spaces/snorkelai/snorkelai_mistral_pairrm_dpo_text_inference)
9
+
10
+ We also provide an inference endpoint for everyone to test the model.
11
+ It may initially take a few minutes to activate, but will eventually operate at the standard speed of HF's 7B model text inference endpoint.
12
+ The speed of inference depends on HF endpoint performance and is not related to Snorkel offerings.
13
+ This endpoint is designed for initial trials, not for ongoing production use. Have fun!
14
+
15
+ ```
16
+ import requests
17
+
18
+ API_URL = "https://t1q6ks6fusyg1qq7.us-east-1.aws.endpoints.huggingface.cloud"
19
+ headers = {
20
+ "Accept" : "application/json",
21
+ "Content-Type": "application/json"
22
+ }
23
+
24
+ def query(payload):
25
+ response = requests.post(API_URL, headers=headers, json=payload)
26
+ return response.json()
27
+
28
+ output = query({
29
+ "inputs": "[INST] Recommend me some Hollywood movies [/INST]",
30
+ "parameters": {}
31
+ })
32
+ ```
33
+
34
+
35
+
36
  ### Dataset:
37
  Training dataset: [snorkelai/Snorkel-Mistral-PairRM-DPO-Dataset](https://huggingface.co/datasets/snorkelai/Snorkel-Mistral-PairRM-DPO-Dataset)
38
 
 
47
  This overview provides a high-level summary of our approach.
48
  We plan to release more detailed results and findings in the coming weeks on the [Snorkel blog.](https://snorkel.ai/blog/)
49
 
50
+ The prompt format follows the Mistral model:
51
+
52
+ ```[INST] {prompt} [/INST]```
53
+
54
  ### Training recipe:
55
  - The provided data is formatted to be compatible with the Hugging Face's [Zephyr recipe](https://github.com/huggingface/alignment-handbook/tree/main/recipes/zephyr-7b-beta).
56
  We executed the n_th DPO iteration using the "train/test_iteration_{n}".
output.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:228dc3a71f4d232fdacef771cc020f0ad5711b34c88d080861d22ef6db4fe9dc
3
  size 3861346868
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7918d9a4ececb4f3a0e9f2703b19161746833a8df0ecf9051c40034a9ffcd3f1
3
  size 3861346868