Universal Assisted Generation: Faster Decoding with Any Assistant Model
ā¢
40
Joao Gante
joaogante
AI & ML interests
None yet
Articles
Organizations
Posts
3
Post
2716
New sampling strategy dropped in š¤ transformers -- Min P sampling š„
Are you tired of having
Min P consists of a dynamic token filter -- as opposed to Top K, which keeps the K most likely tokens, and Top P, which keeps the most likely tokens up to a fixed cumulative probability, both static filters. Min P takes a base probability (defined in the
š High probability token present -> aggressive filter (we don't want to miss on that high-probability case and risk derailing generation)
š No high probability token present -> relaxed filter (there are many continuation possibilities that the model finds plausible)
You should set
Kudos to @kalomaze and @menhguin for creating this technique š„ Read their discussion in the original issue for benchmarks (https://github.com/huggingface/transformers/issues/27670)
Copy-pasteable version of the example in the image below here: https://pastebin.com/VqXNtuxd
Have fun experimenting! š
Are you tired of having
top_k arbitrarily discarding high-quality continuations? Or top_p forgetting to exclude low-probability tokens, derailing your generation? Try out the new min_p flag in generate, fresh from a PR merged today! š„¬Min P consists of a dynamic token filter -- as opposed to Top K, which keeps the K most likely tokens, and Top P, which keeps the most likely tokens up to a fixed cumulative probability, both static filters. Min P takes a base probability (defined in the
min_p flag) and multiplies it by the probability of the most likely token in the distribution for the next token. All tokens less likely than the resulting value are filtered. What happens with this strategy?š High probability token present -> aggressive filter (we don't want to miss on that high-probability case and risk derailing generation)
š No high probability token present -> relaxed filter (there are many continuation possibilities that the model finds plausible)
You should set
min_p to a low value, between 0.05 and 0.1. It behaves particularly well for creative text generation when paired up with temperature > 1.Kudos to @kalomaze and @menhguin for creating this technique š„ Read their discussion in the original issue for benchmarks (https://github.com/huggingface/transformers/issues/27670)
Copy-pasteable version of the example in the image below here: https://pastebin.com/VqXNtuxd
Have fun experimenting! š
Post
2536
Adding a long prompt can help you fight LLM hallucinations. However, if you know exactly how you want your LLM output constrained, there are much better strategies! šŖ
Did you know you can force your LLM to ALWAYS generate a valid JSON file? Or to follow a well-defined answer template? You can do that and more with the š¤ transformers-compatible
It doesn't only allow you to master your LLM -- your text generation application will also become faster! š„ The more constrained your text generation is, the bigger speedups you'll see!
Follow @remi and other
Did you know you can force your LLM to ALWAYS generate a valid JSON file? Or to follow a well-defined answer template? You can do that and more with the š¤ transformers-compatible
outlines library.It doesn't only allow you to master your LLM -- your text generation application will also become faster! š„ The more constrained your text generation is, the bigger speedups you'll see!
Follow @remi and other
outlines folks to stay on top of the constrained generation game š§
spaces
6
models
8
joaogante/dummy_synthid_detector
Updated
ā¢
137
joaogante/tiny-random-gpt2-with-generation-config
Updated
ā¢
71
joaogante/Mistral-7B-Instruct-v0.2-medusa-wikitext
Updated
ā¢
6
joaogante/TinyLlama-1.1B-Chat-v1.0-medusa-wikitext
Updated
ā¢
4
joaogante/Mistral-7B-Instruct-v0.2-medusa-vicuna
Updated
ā¢
2
joaogante/test_audio
Automatic Speech Recognition
ā¢
Updated
ā¢
19
joaogante/test_text
Fill-Mask
ā¢
Updated
ā¢
27
joaogante/test_img
Image Feature Extraction
ā¢
Updated
ā¢
10
ā¢
1