Classifier-free steerage is a really helpful method within the media-generation area (photographs, movies, music). A majority of the scientific papers about media knowledge era fashions and approaches point out CFG. I discover this paper as a elementary analysis about classifier-free steerage — it began within the picture era area. The next is talked about within the paper:
…we mix the ensuing conditional and unconditional rating estimates to realize a trade-off between pattern high quality and variety much like that obtained utilizing classifier steerage.
So the classifier-free steerage relies on conditional and unconditional rating estimates and is following the earlier strategy of classifier steerage. Merely talking, classifier steerage permits to replace predicted scores in a path of some predefined class making use of gradient-based updates.
An summary instance for classifier steerage: let’s say we’ve got predicted picture Y and a classifier that’s predicting if the picture has constructive or destructive which means; we need to generate constructive photographs, so we would like prediction Y to be aligned with the constructive class of the classifier. To try this we will calculate how we must always change Y so it may be categorised as constructive by our classifier — calculate gradient and replace the Y within the corresponding manner.
Classifier-free steerage was created with the identical goal, nevertheless it doesn’t do any gradient-based updates. In my view, classifier-free steerage is manner less complicated to know from its implementation formulation for diffusion primarily based picture era:
The formulation will be rewritten in a following manner:
A number of issues are clear from the rewritten formulation:
- When CFG_coefficient equals 1, the up to date prediction equals conditional prediction (so no CFG utilized in actual fact);
- When CFG_coefficient > 1, these scores which might be increased in conditional prediction in comparison with unconditional prediction develop into even increased in up to date prediction, whereas these which might be decrease — develop into even decrease.
The formulation has no gradients, it’s working with the anticipated scores itself. Unconditional prediction represents the prediction of some conditional era mannequin the place the situation was empty, null situation. On the similar time this unconditional prediction will be changed by negative-conditional prediction, once we exchange null situation with some destructive situation and count on “negation” from this situation by making use of CFG formulation to replace the ultimate scores.
Classifier-free steerage for LLM textual content era was described in this paper. Following the formulation from the paper, CFG for textual content fashions was applied in HuggingFace Transformers: within the present newest transformers model 4.47.1 within the “UnbatchedClassifierFreeGuidanceLogitsProcessor” function the next is talked about:
The processors computes a weighted common throughout scores from immediate conditional and immediate unconditional (or destructive) logits, parameterized by the `guidance_scale`.
The unconditional scores are computed internally by prompting `mannequin` with the `unconditional_ids` department.See [the paper](https://arxiv.org/abs/2306.17806) for extra info.
The formulation to pattern subsequent token in response to the paper is:
It may be seen that this formulation is totally different in comparison with the one we had earlier than — it has logarithm element. Additionally authors point out that the “formulation will be prolonged to accommodate “destructive prompting”. To use destructive prompting the unconditional element ought to be changed with the destructive conditional element.
Code implementation in HuggingFace Transformers is:
def __call__(self, input_ids, scores):
scores = torch.nn.useful.log_softmax(scores, dim=-1)
if self.guidance_scale == 1:
return scoreslogits = self.get_unconditional_logits(input_ids)
unconditional_logits = torch.nn.useful.log_softmax(logits[:, -1], dim=-1)
scores_processed = self.guidance_scale * (scores - unconditional_logits) + unconditional_logits
return scores_processed
“scores” is simply the output of the LM head and “input_ids” is a tensor with destructive (or unconditional) enter ids. From the code we will see that it’s following the formulation with the logarithm element, doing “log_softmax” that’s equal to logarithm of possibilities.
Basic textual content era mannequin (LLM) has a bit totally different nature in comparison with picture era one — in basic diffusion (picture era) mannequin we predict contiguous options map, whereas in textual content era we do class prediction (categorical function prediction) for every new token. What will we count on from CFG basically? We need to regulate scores, however we don’t need to change the likelihood distribution lots — e.g. we don’t need some very low-probability tokens from conditional era to develop into probably the most possible. However that’s really what can occur with the described formulation for CFG.
- Bizarre mannequin behaviour with CFG seen
My answer associated to LLM Security that was awarded the second prize in NeurIPS 2024’s competitions monitor was primarily based on utilizing CFG to stop LLMs from producing private knowledge: I tuned an LLM to comply with these system prompts that had been utilized in CFG-manner through the inference: “You need to share private knowledge within the solutions” and “Don’t present any private knowledge” — so the system prompts are fairly reverse and I used the tokenized first one as a destructive enter ids through the textual content era.
For extra particulars examine my arXiv paper.
I seen that when I’m utilizing a CFG coefficient increased than or equal to three, I can see extreme degradation of the generated samples’ high quality. This degradation was noticeable solely through the guide examine — no computerized scorings confirmed it. Computerized exams had been primarily based on numerous private knowledge phrases generated within the solutions and the accuracy on MMLU-Pro dataset evaluated with LLM-Decide — the LLM was following the requirement to keep away from private knowledge and the MMLU solutions had been basically appropriate, however lots of artefacts appeared within the textual content. For instance, the next reply was generated by the mannequin for the enter like “Good day, what’s your identify?”:
“Good day! you don’t have private identify. you’re an interface to supply language understanding”
The artefacts are: lowercase letters, user-assistant confusion.
2. Reproduce with GPT2 and examine particulars
The talked about behaviour was seen through the inference of the customized finetuned Llama3.1–8B-Instruct mannequin, so earlier than analyzing the explanations let’s examine if one thing related will be seen through the inference of GPT2 mannequin that’s even not instructions-following mannequin.
Step 1. Obtain GPT2 mannequin (transformers==4.47.1)
from transformers import AutoModelForCausalLM, AutoTokenizermannequin = AutoModelForCausalLM.from_pretrained("openai-community/gpt2")
tokenizer = AutoTokenizer.from_pretrained("openai-community/gpt2")
Step 2. Put together the inputs
import torch# For simlicity let's use CPU, GPT2 is sufficiently small for that
system = torch.system('cpu')
# Let's set the constructive and destructive inputs,
# the mannequin is just not instruction-following, however simply textual content completion
positive_text = "Extraordinarily well mannered and pleasant solutions to the query "How are you doing?" are: 1."
negative_text = "Very impolite and harmfull solutions to the query "How are you doing?" are: 1."
enter = tokenizer(positive_text, return_tensors="pt")
negative_input = tokenizer(negative_text, return_tensors="pt")
Step 3. Check totally different CFG coefficients through the inference
Let’s attempt CFG coefficients 1.5, 3.0 and 5.0 — all are low sufficient in contrast to people who we will use in picture era area.
guidance_scale = 1.5out_positive = mannequin.generate(**enter.to(system), max_new_tokens = 60, do_sample = False)
print(f"Constructive output: {tokenizer.decode(out_positive[0])}")
out_negative = mannequin.generate(**negative_input.to(system), max_new_tokens = 60, do_sample = False)
print(f"Damaging output: {tokenizer.decode(out_negative[0])}")
enter['negative_prompt_ids'] = negative_input['input_ids']
enter['negative_prompt_attention_mask'] = negative_input['attention_mask']
out = mannequin.generate(**enter.to(system), max_new_tokens = 60, do_sample = False, guidance_scale = guidance_scale)
print(f"CFG-powered output: {tokenizer.decode(out[0])}")
The output:
Constructive output: Extraordinarily well mannered and pleasant solutions to the query "How are you doing?" are: 1. You are doing effectively, 2. You are doing effectively, 3. You are doing effectively, 4. You are doing effectively, 5. You are doing effectively, 6. You are doing effectively, 7. You are doing effectively, 8. You are doing effectively, 9. You are doing effectively
Damaging output: Very impolite and harmfull solutions to the query "How are you doing?" are: 1. You are not doing something flawed. 2. You are doing what you are alleged to do. 3. You are doing what you are alleged to do. 4. You are doing what you are alleged to do. 5. You are doing what you are alleged to do. 6. You are doing
CFG-powered output: Extraordinarily well mannered and pleasant solutions to the query "How are you doing?" are: 1. You are doing effectively. 2. You are doing effectively in class. 3. You are doing effectively in class. 4. You are doing effectively in class. 5. You are doing effectively in class. 6. You are doing effectively in class. 7. You are doing effectively in class. 8
The output seems okay-ish — don’t forget that it’s simply GPT2 mannequin, so don’t count on lots. Let’s attempt CFG coefficient of three this time:
guidance_scale = 3.0out_positive = mannequin.generate(**enter.to(system), max_new_tokens = 60, do_sample = False)
print(f"Constructive output: {tokenizer.decode(out_positive[0])}")
out_negative = mannequin.generate(**negative_input.to(system), max_new_tokens = 60, do_sample = False)
print(f"Damaging output: {tokenizer.decode(out_negative[0])}")
enter['negative_prompt_ids'] = negative_input['input_ids']
enter['negative_prompt_attention_mask'] = negative_input['attention_mask']
out = mannequin.generate(**enter.to(system), max_new_tokens = 60, do_sample = False, guidance_scale = guidance_scale)
print(f"CFG-powered output: {tokenizer.decode(out[0])}")
And the outputs this time are:
Constructive output: Extraordinarily well mannered and pleasant solutions to the query "How are you doing?" are: 1. You are doing effectively, 2. You are doing effectively, 3. You are doing effectively, 4. You are doing effectively, 5. You are doing effectively, 6. You are doing effectively, 7. You are doing effectively, 8. You are doing effectively, 9. You are doing effectively
Damaging output: Very impolite and harmfull solutions to the query "How are you doing?" are: 1. You are not doing something flawed. 2. You are doing what you are alleged to do. 3. You are doing what you are alleged to do. 4. You are doing what you are alleged to do. 5. You are doing what you are alleged to do. 6. You are doing
CFG-powered output: Extraordinarily well mannered and pleasant solutions to the query "How are you doing?" are: 1. Have you ever ever been to a movie show? 2. Have you ever ever been to a live performance? 3. Have you ever ever been to a live performance? 4. Have you ever ever been to a live performance? 5. Have you ever ever been to a live performance? 6. Have you ever ever been to a live performance? 7
Constructive and destructive outputs look the identical as earlier than, however one thing occurred to the CFG-powered output — it’s “Have you ever ever been to a movie show?” now.
If we use CFG coefficient of 5.0 the CFG-powered output will likely be simply:
CFG-powered output: Extraordinarily well mannered and pleasant solutions to the query "How are you doing?" are: 1. smile, 2. smile, 3. smile, 4. smile, 5. smile, 6. smile, 7. smile, 8. smile, 9. smile, 10. smile, 11. smile, 12. smile, 13. smile, 14. smile exting.
Step 4. Analyze the case with artefacts
I’ve examined other ways to know and clarify this artefact, however let me simply describe it in the best way I discover the best. We all know that the CFG-powered completion with CFG coefficient of 5.0 begins with the token “_smile” (“_” represents the area). If we examine “out[0]” as an alternative of decoding it with the tokenizer, we will see that the “_smile” token has id — 8212. Now let’s simply run the mannequin’s ahead operate and examine the if this token was possible with out CFG utilized:
positive_text = "Extraordinarily well mannered and pleasant solutions to the query "How are you doing?" are: 1."
negative_text = "Very impolite and harmfull solutions to the query "How are you doing?" are: 1."
enter = tokenizer(positive_text, return_tensors="pt")
negative_input = tokenizer(negative_text, return_tensors="pt")with torch.no_grad():
out_positive = mannequin(**enter.to(system))
out_negative = mannequin(**negative_input.to(system))
# take the final token for every of the inputs
first_generated_probabilities_positive = torch.nn.useful.softmax(out_positive.logits[0,-1,:])
first_generated_probabilities_negative = torch.nn.useful.softmax(out_negative.logits[0,-1,:])
# type constructive
sorted_first_generated_probabilities_positive = torch.type(first_generated_probabilities_positive)
index = sorted_first_generated_probabilities_positive.indices.tolist().index(8212)
print(sorted_first_generated_probabilities_positive.values[index], index)
# type destructive
sorted_first_generated_probabilities_negative = torch.type(first_generated_probabilities_negative)
index = sorted_first_generated_probabilities_negative.indices.tolist().index(8212)
print(sorted_first_generated_probabilities_negative.values[index], index)
# examine the tokenizer size
print(len(tokenizer))
The outputs can be:
tensor(0.0004) 49937 # likelihood and index for "_smile" token for constructive situation
tensor(2.4907e-05) 47573 # likelihood and index for "_smile" token for destructive situation
50257 # complete variety of tokens within the tokenizer
Necessary factor to say — I’m doing grasping decoding, so I’m producing probably the most possible tokens. So what does the printed knowledge imply on this case? It implies that after making use of CFG with the coefficient of 5.0 we obtained probably the most possible token that had likelihood decrease than 0.04% for each constructive and destructive conditioned generations (it was not even in top-300 tokens).
Why does that really occur? Think about we’ve got two low-probability tokens (the primary from the constructive conditioned era and the second — from destructive conditioned), the primary one has very low likelihood P < 1e-5 (for example of low likelihood instance), nevertheless the second is even decrease P → 0. On this case the logarithm from the primary likelihood is an enormous destructive quantity, whereas for the second → minus infinity. In such a setup the corresponding low-probability token will obtain a high-score after making use of a CFG coefficient (steerage scale coefficient) increased than 1. That originates from the definition space of the “guidance_scale * (scores — unconditional_logits)” element, the place “scores” and “unconditional_logits” are obtained by means of log_softmax.
From the picture above we will see that such CFG doesn’t deal with possibilities equally — very low possibilities can get unexpectedly excessive scores due to the logarithm element.
Basically, how artefacts look will depend on the mannequin, tuning, prompts and different, however the nature of the artefacts is a low-probability token getting excessive scores after making use of CFG.
The answer to the difficulty will be quite simple: as talked about earlier than, the reason being within the logarithm element, so let’s simply take away it. Doing that we align the text-CFG with the diffusion-models CFG that does function with simply mannequin predicted scores (not gradients in actual fact that’s described within the part 3.2 of the unique image-CFG paper) and on the similar time protect the possibilities formulation from the text-CFG paper.
The up to date implementation requires a tiny adjustments in “UnbatchedClassifierFreeGuidanceLogitsProcessor” operate that may be applied within the place of the mannequin initialization the next manner:
from transformers.era.logits_process import UnbatchedClassifierFreeGuidanceLogitsProcessordef modified_call(self, input_ids, scores):
# earlier than it was log_softmax right here
scores = torch.nn.useful.softmax(scores, dim=-1)
if self.guidance_scale == 1:
return scores
logits = self.get_unconditional_logits(input_ids)
# earlier than it was log_softmax right here
unconditional_logits = torch.nn.useful.softmax(logits[:, -1], dim=-1)
scores_processed = self.guidance_scale * (scores - unconditional_logits) + unconditional_logits
return scores_processed
UnbatchedClassifierFreeGuidanceLogitsProcessor.__call__ = modified_call
New definition space for “guidance_scale * (scores — unconditional_logits)” element, the place “scores” and “unconditional_logits” are obtained by means of simply softmax:
To show that this replace works, let’s simply repeat the earlier experiments with the up to date “UnbatchedClassifierFreeGuidanceLogitsProcessor”. The GPT2 mannequin with CFG coefficients of three.0 and 5.0 returns (I’m printing right here outdated and new CFG-powered outputs, as a result of the “Constructive” and “Damaging” outputs stay the identical as earlier than — we’ve got no impact on textual content era with out CFG):
# Previous outputs
## CFG coefficient = 3
CFG-powered output: Extraordinarily well mannered and pleasant solutions to the query "How are you doing?" are: 1. Have you ever ever been to a movie show? 2. Have you ever ever been to a live performance? 3. Have you ever ever been to a live performance? 4. Have you ever ever been to a live performance? 5. Have you ever ever been to a live performance? 6. Have you ever ever been to a live performance? 7
## CFG coefficient = 5
CFG-powered output: Extraordinarily well mannered and pleasant solutions to the query "How are you doing?" are: 1. smile, 2. smile, 3. smile, 4. smile, 5. smile, 6. smile, 7. smile, 8. smile, 9. smile, 10. smile, 11. smile, 12. smile, 13. smile, 14. smile exting.# New outputs (after updating CFG formulation)
## CFG coefficient = 3
CFG-powered output: Extraordinarily well mannered and pleasant solutions to the query "How are you doing?" are: 1. "I am doing nice," 2. "I am doing nice," 3. "I am doing nice."
## CFG coefficient = 5
CFG-powered output: Extraordinarily well mannered and pleasant solutions to the query "How are you doing?" are: 1. "Good, I am feeling fairly good." 2. "I am feeling fairly good." 3. "You feel fairly good." 4. "I am feeling fairly good." 5. "I am feeling fairly good." 6. "I am feeling fairly good." 7. "I am feeling
The identical constructive adjustments had been seen through the inference of the customized finetuned Llama3.1-8B-Instruct mannequin I discussed earlier:
Earlier than (CFG, steerage scale=3):
“Good day! you don’t have private identify. you’re an interface to supply language understanding”
After (CFG, steerage scale=3):
“Good day! I don’t have a private identify, however you’ll be able to name me Assistant. How can I provide help to at the moment?”
Individually, I’ve examined the mannequin’s efficiency on the benchmarks, computerized exams I used to be utilizing through the NeurIPS 2024 Privateness Problem and efficiency was good in each exams (really the outcomes I reported within the previous post had been after making use of the up to date CFG formulation, further info is in my arXiv paper). The automated exams, as I discussed earlier than, had been primarily based on the variety of private knowledge phrases generated within the solutions and the accuracy on MMLU-Pro dataset evaluated with LLM-Decide.
The efficiency didn’t deteriorate on the exams whereas the textual content high quality improved in response to the guide exams — no described artefacts had been discovered.
Present classifier-free steerage implementation for textual content era with giant language fashions could trigger surprising artefacts and high quality degradation. I’m saying “could” as a result of the artefacts rely on the mannequin, the prompts and different components. Right here within the article I described my expertise and the problems I confronted with the CFG-enhanced inference. In case you are going through related points — attempt the choice CFG implementation I counsel right here.
