MLPSpeculator. (#1865)

# What does this PR do?

<!--
Congratulations! You've made it this far! You're not quite done yet
though.

Once merged, your PR is going to appear in the release notes with the
title you set, so make sure it's a great title that fully reflects the
extent of your awesome contribution.

Then, please replace this with a description of the change and which
issue is fixed (if applicable). Please also include relevant motivation
and context. List any dependencies (if any) that are required for this
change.

Once you're done, someone will review your PR shortly (see the section
"Who can review?" below to tag some potential reviewers). They may
suggest changes to make the code even better. If no one reviewed your PR
after a week has passed, don't hesitate to post a new comment
@-mentioning the same persons---sometimes notifications get lost.
-->

<!-- Remove if not applicable -->

Fixes # (issue)


## Before submitting
- [ ] This PR fixes a typo or improves the docs (you can dismiss the
other checks if that's the case).
- [ ] Did you read the [contributor
guideline](https://github.com/huggingface/transformers/blob/main/CONTRIBUTING.md#start-contributing-pull-requests),
      Pull Request section?
- [ ] Was this discussed/approved via a Github issue or the
[forum](https://discuss.huggingface.co/)? Please add a link
      to it if that's the case.
- [ ] Did you make sure to update the documentation with your changes?
Here are the
[documentation
guidelines](https://github.com/huggingface/transformers/tree/main/docs),
and
[here are tips on formatting
docstrings](https://github.com/huggingface/transformers/tree/main/docs#writing-source-documentation

).
- [ ] Did you write any new necessary tests?


## Who can review?

Anyone in the community is free to review the PR once the tests have
passed. Feel free to tag
members/contributors who may be interested in your PR.

<!-- Your PR will be replied to more quickly if you can figure out the
right person to tag with @


@OlivierDehaene OR @Narsil

 -->

---------
Co-authored-by: Joshua Rosenkranz <joshua.rosenkranz@gmail.com>

server/text_generation_server/models/flash_santacoder.py

@@ -29,7 +29,7 @@ class FlashSantacoderSharded(FlashCausalLM):
         model_id: str,
         revision: Optional[str] = None,
         quantize: Optional[str] = None,
-        use_medusa: Optional[str] = None,
+        speculator: Optional[str] = None,
         dtype: Optional[torch.dtype] = None,
         trust_remote_code: bool = False,
     ):
@@ -57,7 +57,7 @@ class FlashSantacoderSharded(FlashCausalLM):
             trust_remote_code=True,
         )
         config.quantize = quantize
-        config.use_medusa = use_medusa
+        config.speculator = speculator
         config.transpose = config.architectures[0].startswith("GPT2")
 
         torch.distributed.barrier(group=self.process_group)

server/text_generation_server/models/flash_starcoder2.py

@@ -29,7 +29,7 @@ class FlashStarcoder2(BaseFlashMistral):
         model_id: str,
         revision: Optional[str] = None,
         quantize: Optional[str] = None,
-        use_medusa: Optional[str] = None,
+        speculator: Optional[str] = None,
         dtype: Optional[torch.dtype] = None,
         trust_remote_code: bool = False,
     ):
@@ -52,7 +52,7 @@ class FlashStarcoder2(BaseFlashMistral):
             model_id, revision=revision, trust_remote_code=trust_remote_code
         )
         config.quantize = quantize
-        config.use_medusa = use_medusa
+        config.speculator = speculator
 
         # Set context windows
         if config.sliding_window is not None:

server/text_generation_server/models/galactica.py

@@ -167,7 +167,7 @@ class GalacticaSharded(CausalLM):
         model_id: str,
         revision: Optional[str] = None,
         quantize: Optional[str] = None,
-        use_medusa: Optional[str] = None,
+        speculator: Optional[str] = None,
         dtype: Optional[torch.dtype] = None,
         trust_remote_code: bool = False,
     ):
@@ -195,7 +195,7 @@ class GalacticaSharded(CausalLM):
         )
         config.quantize = quantize
         tokenizer.pad_token_id = config.pad_token_id
-        config.use_medusa = use_medusa
+        config.speculator = speculator
 
         torch.distributed.barrier(group=self.process_group)
         filenames = weight_files(model_id, revision=revision, extension=".safetensors")

server/text_generation_server/models/gpt_neox.py

@@ -24,7 +24,7 @@ class GPTNeoxSharded(CausalLM):
         model_id: str,
         revision: Optional[str] = None,
         quantize: Optional[str] = None,
-        use_medusa: Optional[str] = None,
+        speculator: Optional[str] = None,
         dtype: Optional[torch.dtype] = None,
         trust_remote_code: bool = False,
     ):
@@ -51,7 +51,7 @@ class GPTNeoxSharded(CausalLM):
             trust_remote_code=trust_remote_code,
         )
         config.quantize = quantize
-        config.use_medusa = use_medusa
+        config.speculator = speculator
 
         torch.distributed.barrier(group=self.process_group)
         filenames = weight_files(model_id, revision=revision, extension=".safetensors")

server/text_generation_server/models/idefics.py

@@ -31,7 +31,7 @@ class IDEFICSSharded(IdeficsCausalLM):
         model_id: str,
         revision: Optional[str] = None,
         quantize: Optional[str] = None,
-        use_medusa: Optional[str] = None,
+        speculator: Optional[str] = None,
         dtype: Optional[torch.dtype] = None,
         trust_remote_code: bool = False,
     ):
@@ -52,7 +52,7 @@ class IDEFICSSharded(IdeficsCausalLM):
             trust_remote_code=trust_remote_code,
         )
         config.quantize = quantize
-        config.use_medusa = use_medusa
+        config.speculator = speculator
         config.vision_config.quantize = quantize
 
         tokenizer = LlamaTokenizerFast.from_pretrained(

server/text_generation_server/models/idefics2.py

@@ -18,7 +18,7 @@ class Idefics2(VlmCausalLM):
         model_id: str,
         revision: Optional[str] = None,
         quantize: Optional[str] = None,
-        use_medusa: Optional[str] = None,
+        speculator: Optional[str] = None,
         dtype: Optional[torch.dtype] = None,
         trust_remote_code: bool = False,
     ):
@@ -35,7 +35,7 @@ class Idefics2(VlmCausalLM):
             model_id=model_id,
             revision=revision,
             quantize=quantize,
-            use_medusa=use_medusa,
+            speculator=speculator,
             dtype=dtype,
             trust_remote_code=trust_remote_code,
         )

server/text_generation_server/models/llava_next.py

@@ -18,7 +18,7 @@ class LlavaNext(VlmCausalLM):
         model_id: str,
         revision: Optional[str] = None,
         quantize: Optional[str] = None,
-        use_medusa: Optional[str] = None,
+        speculator: Optional[str] = None,
         dtype: Optional[torch.dtype] = None,
         trust_remote_code: bool = False,
     ):
@@ -30,7 +30,7 @@ class LlavaNext(VlmCausalLM):
             model_id=model_id,
             revision=revision,
             quantize=quantize,
-            use_medusa=use_medusa,
+            speculator=speculator,
             dtype=dtype,
             trust_remote_code=trust_remote_code,
         )

server/text_generation_server/models/mamba.py

@@ -408,7 +408,7 @@ class Mamba(Model):
         model_id: str,
         revision: Optional[str] = None,
         quantize: Optional[str] = None,
-        use_medusa: Optional[str] = None,
+        speculator: Optional[str] = None,
         dtype: Optional[torch.dtype] = None,
         trust_remote_code: bool = False,
     ):
@@ -445,7 +445,7 @@ class Mamba(Model):
         tokenizer.pad_token = tokenizer.eos_token
 
         config.quantize = quantize
-        config.use_medusa = use_medusa
+        config.speculator = speculator
         torch.distributed.barrier(group=self.process_group)
         filenames = weight_files(model_id, revision=revision, extension=".safetensors")
         weights = Weights(filenames, device, dtype, process_group=self.process_group)

server/text_generation_server/models/mpt.py

@@ -43,7 +43,7 @@ class MPTSharded(CausalLM):
         model_id: str,
         revision: Optional[str] = None,
         quantize: Optional[str] = None,
-        use_medusa: Optional[str] = None,
+        speculator: Optional[str] = None,
         dtype: Optional[torch.dtype] = None,
         trust_remote_code: bool = False,
     ):
@@ -76,7 +76,7 @@ class MPTSharded(CausalLM):
             config = json.load(f)
         config = PretrainedConfig(**config)
         config.quantize = quantize
-        config.use_medusa = use_medusa
+        config.speculator = speculator
 
         torch.distributed.barrier(group=self.process_group)
 

server/text_generation_server/models/opt.py

@@ -22,7 +22,7 @@ class OPTSharded(CausalLM):
         model_id: str,
         revision: Optional[str] = None,
         quantize: Optional[str] = None,
-        use_medusa: Optional[str] = None,
+        speculator: Optional[str] = None,
         dtype: Optional[torch.dtype] = None,
         trust_remote_code: bool = False,
     ):
@@ -48,7 +48,7 @@ class OPTSharded(CausalLM):
             trust_remote_code=trust_remote_code,
         )
         config.quantize = quantize
-        config.use_medusa = use_medusa
+        config.speculator = speculator
         tokenizer.pad_token_id = config.pad_token_id
 
         torch.distributed.barrier(group=self.process_group)

server/text_generation_server/models/phi.py

@@ -22,7 +22,7 @@ class Phi(CausalLM):
         model_id: str,
         revision: Optional[str] = None,
         quantize: Optional[str] = None,
-        use_medusa: Optional[str] = None,
+        speculator: Optional[str] = None,
         dtype: Optional[torch.dtype] = None,
         trust_remote_code: bool = False,
     ):
@@ -53,7 +53,7 @@ class Phi(CausalLM):
         tokenizer.pad_token = tokenizer.eos_token
 
         config.quantize = quantize
-        config.use_medusa = use_medusa
+        config.speculator = speculator
         torch.distributed.barrier(group=self.process_group)
         filenames = weight_files(model_id, revision=revision, extension=".safetensors")
         weights = Weights(filenames, device, dtype, process_group=self.process_group)

server/text_generation_server/models/rw.py

@@ -12,11 +12,11 @@ class RW(CausalLM):
         model_id: str,
         revision: Optional[str] = None,
         quantize: Optional[str] = None,
-        use_medusa: Optional[str] = None,
+        speculator: Optional[str] = None,
         dtype: Optional[torch.dtype] = None,
         trust_remote_code: bool = False,
     ):
-        if use_medusa:
+        if speculator:
             raise RuntimeError("Medusa decoding is not enabled for AutoModel")
 
         if torch.cuda.is_available():

server/text_generation_server/models/santacoder.py

@@ -19,7 +19,7 @@ class SantaCoder(CausalLM):
         model_id: str,
         revision: Optional[str] = None,
         quantize: Optional[str] = None,
-        use_medusa: Optional[str] = None,
+        speculator: Optional[str] = None,
         dtype: Optional[torch.dtype] = None,
         trust_remote_code: bool = False,
     ):

server/text_generation_server/models/seq2seq_lm.py

@@ -532,12 +532,12 @@ class Seq2SeqLM(Model):
         model_id: str,
         revision: Optional[str] = None,
         quantize: Optional[str] = None,
-        use_medusa: Optional[str] = None,
+        speculator: Optional[str] = None,
         dtype: Optional[torch.dtype] = None,
         trust_remote_code: bool = False,
     ):
-        if use_medusa:
-            raise RuntimeError("Medusa decoding is not enabled for AutoModel")
+        if speculator:
+            raise RuntimeError("Speculator decoding is not enabled for AutoModel")
 
         if torch.cuda.is_available():
             device = torch.device("cuda")

server/text_generation_server/models/t5.py

@@ -25,7 +25,7 @@ class T5Sharded(Seq2SeqLM):
         model_id: str,
         revision: Optional[str] = None,
         quantize: Optional[str] = None,
-        use_medusa: Optional[str] = None,
+        speculator: Optional[str] = None,
         dtype: Optional[torch.dtype] = None,
         trust_remote_code: bool = False,
     ):
@@ -43,7 +43,7 @@ class T5Sharded(Seq2SeqLM):
             trust_remote_code=trust_remote_code,
         )
         config.quantize = quantize
-        config.use_medusa = use_medusa
+        config.speculator = speculator
 
         tokenizer = AutoTokenizer.from_pretrained(
             model_id,