Fix issues in sample code: Invalid reward_tokenizer and issue in returning scores
Browse filesCurrently, when running the code, following errors occur. I fixed the two.
README.md
CHANGED
|
@@ -92,11 +92,10 @@ class GPTRewardModel(nn.Module):
|
|
| 92 |
c_ind = c_inds[0].item() if len(c_inds) > 0 else input_ids.shape[1]
|
| 93 |
scores.append(rewards[i, c_ind - 1])
|
| 94 |
return scores
|
| 95 |
-
return scores
|
| 96 |
|
| 97 |
## Load the model and tokenizer
|
| 98 |
|
| 99 |
-
reward_model = GPTRewardModel("meta-llama/Llama-2-7b-chat-hf"
|
| 100 |
reward_tokenizer = reward_model.tokenizer
|
| 101 |
reward_tokenizer.truncation_side = "left"
|
| 102 |
|
|
|
|
| 92 |
c_ind = c_inds[0].item() if len(c_inds) > 0 else input_ids.shape[1]
|
| 93 |
scores.append(rewards[i, c_ind - 1])
|
| 94 |
return scores
|
|
|
|
| 95 |
|
| 96 |
## Load the model and tokenizer
|
| 97 |
|
| 98 |
+
reward_model = GPTRewardModel("meta-llama/Llama-2-7b-chat-hf")
|
| 99 |
reward_tokenizer = reward_model.tokenizer
|
| 100 |
reward_tokenizer.truncation_side = "left"
|
| 101 |
|