Skip to content

Commit 64c01b8

Browse files
committed
Updated special token description
1 parent f07cf1e commit 64c01b8

File tree

1 file changed

+17
-3
lines changed

1 file changed

+17
-3
lines changed

README.md

Lines changed: 17 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -7,9 +7,9 @@ The 'llama-recipes' repository is a companion to the [Meta Llama 2](https://gith
77
> | Token | Description |
88
> |---|---|
99
> `<\|begin_of_text\|>` | This is equivalent to the BOS token. |
10-
> `<\|eot_id\|>` | This signifies the end of the message in a turn. |
10+
> `<\|eot_id\|>` | This signifies the end of the message in a turn. The generate function needs to be set up as shown below or in [this example](./recipes/inference/local_inference/chat_completion/chat_completion.py) to terminate the generation after the turn.|
1111
> `<\|start_header_id\|>{role}<\|end_header_id\|>` | These tokens enclose the role for a particular message. The possible roles can be: system, user, assistant. |
12-
> `<\|end_of_text\|>` | This is equivalent to the EOS token. On generating this token, Llama 3 will cease to generate more tokens |
12+
> `<\|end_of_text\|>` | This is equivalent to the EOS token. Its usually not used during multiturn-conversations. Instead, each message is terminated with `<\|eot_id\|>` |
1313
>
1414
> A multiturn-conversation with Llama 3 follows this prompt template:
1515
> ```
@@ -23,7 +23,21 @@ The 'llama-recipes' repository is a companion to the [Meta Llama 2](https://gith
2323
>
2424
> {{ user_message_2 }}<|eot_id|><|start_header_id|>assistant<|end_header_id|>
2525
> ```
26-
> More details on the new tokenizer and prompt template: <PLACEHOLDER_URL>
26+
>
27+
> To signal the end of the current message the model emits the `<\|eot_id\|>` token. To terminate the generation we need to call the model's generate function as follows:
28+
> ```
29+
> terminators = [
30+
> tokenizer.eos_token_id,
31+
> tokenizer.convert_tokens_to_ids("<|eot_id|>")
32+
> ]
33+
> ...
34+
> outputs = model.generate(
35+
> ...
36+
> eos_token_id=terminators,
37+
> )
38+
> ```
39+
>
40+
> More details on the new tokenizer and prompt template: https://llama.meta.com/docs/model-cards-and-prompt-formats/meta-llama-3#special-tokens-used-with-meta-llama-3
2741
> [!NOTE]
2842
> The llama-recipes repository was recently refactored to promote a better developer experience of using the examples. Some files have been moved to new locations. The `src/` folder has NOT been modified, so the functionality of this repo and package is not impacted.
2943
>

0 commit comments

Comments
 (0)