-
Notifications
You must be signed in to change notification settings - Fork 72
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Support MistralModel and MistralForCausalLM
- Loading branch information
1 parent
b57b660
commit bf09440
Showing
3 changed files
with
83 additions
and
8 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Empty file.
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,78 @@ | ||
| """ | ||
| Each modeling file in this library is a mapping between | ||
| abstract naming of intervention anchor points and actual | ||
| model module defined in the huggingface library. | ||
| We also want to let the intervention library know how to | ||
| config the dimensions of intervention based on model config | ||
| defined in the huggingface library. | ||
| """ | ||
|
|
||
|
|
||
| import torch | ||
| from ..constants import * | ||
|
|
||
|
|
||
| mistral_type_to_module_mapping = { | ||
| "block_input": ("layers[%s]", CONST_INPUT_HOOK), | ||
| "block_output": ("layers[%s]", CONST_OUTPUT_HOOK), | ||
| "mlp_activation": ("layers[%s].mlp.act_fn", CONST_OUTPUT_HOOK), | ||
| "mlp_output": ("layers[%s].mlp", CONST_OUTPUT_HOOK), | ||
| "mlp_input": ("layers[%s].mlp", CONST_INPUT_HOOK), | ||
| "attention_value_output": ("layers[%s].self_attn.o_proj", CONST_INPUT_HOOK), | ||
| "head_attention_value_output": ("layers[%s].self_attn.o_proj", CONST_INPUT_HOOK), | ||
| "attention_output": ("layers[%s].self_attn", CONST_OUTPUT_HOOK), | ||
| "attention_input": ("layers[%s].self_attn", CONST_INPUT_HOOK), | ||
| "query_output": ("layers[%s].self_attn.q_proj", CONST_OUTPUT_HOOK), | ||
| "key_output": ("layers[%s].self_attn.k_proj", CONST_OUTPUT_HOOK), | ||
| "value_output": ("layers[%s].self_attn.v_proj", CONST_OUTPUT_HOOK), | ||
| "head_query_output": ("layers[%s].self_attn.q_proj", CONST_OUTPUT_HOOK), | ||
| "head_key_output": ("layers[%s].self_attn.k_proj", CONST_OUTPUT_HOOK), | ||
| "head_value_output": ("layers[%s].self_attn.v_proj", CONST_OUTPUT_HOOK), | ||
| } | ||
|
|
||
|
|
||
| mistral_type_to_dimension_mapping = { | ||
| "block_input": ("hidden_size",), | ||
| "block_output": ("hidden_size",), | ||
| "mlp_activation": ("intermediate_size",), | ||
| "mlp_output": ("hidden_size",), | ||
| "mlp_input": ("hidden_size",), | ||
| "attention_value_output": ("hidden_size",), | ||
| "head_attention_value_output": ("hidden_size/num_attention_heads",), | ||
| "attention_output": ("hidden_size",), | ||
| "attention_input": ("hidden_size",), | ||
| "query_output": ("hidden_size",), | ||
| "key_output": ("hidden_size",), | ||
| "value_output": ("hidden_size",), | ||
| "head_query_output": ("hidden_size/num_attention_heads",), | ||
| "head_key_output": ("hidden_size/num_attention_heads",), | ||
| "head_value_output": ("hidden_size/num_attention_heads",), | ||
| } | ||
|
|
||
|
|
||
| """llama model with LM head""" | ||
| mistral_lm_type_to_module_mapping = {} | ||
| for k, v in mistral_type_to_module_mapping.items(): | ||
| mistral_lm_type_to_module_mapping[k] = (f"model.{v[0]}", v[1]) | ||
|
|
||
|
|
||
| mistral_lm_type_to_dimension_mapping = mistral_type_to_dimension_mapping | ||
|
|
||
|
|
||
| def create_mistral( | ||
| name="mistralai/Mistral-7B-v0.1", cache_dir=None | ||
| ): | ||
| """Creates a Mistral Causal LM model, config, and tokenizer from the given name and revision""" | ||
| from transformers import AutoModelForCausalLM, AutoTokenizer, AutoConfig | ||
|
|
||
| config = AutoConfig.from_pretrained(name, cache_dir=cache_dir) | ||
| tokenizer = AutoTokenizer.from_pretrained(name, cache_dir=cache_dir) | ||
| llama = AutoModelForCausalLM.from_pretrained( | ||
| name, | ||
| config=config, | ||
| cache_dir=cache_dir, | ||
| torch_dtype=torch.bfloat16, # save memory | ||
| ) | ||
| print("loaded model") | ||
| return config, tokenizer, llama |
bf09440There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Could you please add back the support for GPT2ForSequenceClassification?
bf09440There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thanks for raising this. Ticket created: #138; will resolve in 30 mins.
bf09440There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
resolved! please pull.