Merge pull request #199 from atticusg/atticusDev

Add Qwen model

pyvene/models/intervenable_modelcard.py

@@ -12,6 +12,7 @@
 from .blip.modelings_intervenable_blip_itm import *
 from .backpack_gpt2.modelings_intervenable_backpack_gpt2 import *
 from .llava.modelings_intervenable_llava import *
+from .qwen2.modelings_intervenable_qwen2 import * 
 from .olmo.modelings_intervenable_olmo import *
 
 #########################################################################
@@ -71,7 +72,9 @@
     GRULMHeadModel: gru_lm_type_to_module_mapping,
     GRUForClassification: gru_classifier_type_to_module_mapping,
     BackpackGPT2LMHeadModel: backpack_gpt2_lm_type_to_module_mapping,
-    # new model type goes here after defining the model files
+    hf_models.qwen2.modeling_qwen2.Qwen2Model: qwen2_type_to_module_mapping,
+    hf_models.qwen2.modeling_qwen2.Qwen2ForCausalLM: qwen2_lm_type_to_module_mapping,
+    hf_models.qwen2.modeling_qwen2.Qwen2ForSequenceClassification: qwen2_classifier_type_to_module_mapping,
 }
 if enable_blip:
     type_to_module_mapping[BlipWrapper] = blip_wrapper_type_to_module_mapping
@@ -106,8 +109,11 @@
     GRULMHeadModel: gru_lm_type_to_dimension_mapping,
     GRUForClassification: gru_classifier_type_to_dimension_mapping,
     BackpackGPT2LMHeadModel: backpack_gpt2_lm_type_to_dimension_mapping,
-    # new model type goes here after defining the model files
+    hf_models.qwen2.modeling_qwen2.Qwen2Model: qwen2_type_to_dimension_mapping,
+    hf_models.qwen2.modeling_qwen2.Qwen2ForCausalLM: qwen2_lm_type_to_dimension_mapping,
+    hf_models.qwen2.modeling_qwen2.Qwen2ForSequenceClassification: qwen2_classifier_type_to_dimension_mapping,
 }
+
 if enable_blip:
     type_to_dimension_mapping[BlipWrapper] = blip_wrapper_type_to_dimension_mapping
     type_to_dimension_mapping[BlipITMWrapper] = blip_itm_wrapper_type_to_dimension_mapping

pyvene/models/qwen2/modelings_intervenable_qwen2.py

@@ -0,0 +1,77 @@
+"""
+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 *
+
+qwen2_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, (split_head_and_permute, "n_head")),
+    "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, (split_head_and_permute, "n_head")),
+    "head_key_output": ("layers[%s].self_attn.k_proj", CONST_OUTPUT_HOOK, (split_head_and_permute, "n_kv_head")),
+    "head_value_output": ("layers[%s].self_attn.v_proj", CONST_OUTPUT_HOOK, (split_head_and_permute, "n_kv_head")),
+}
+
+qwen2_type_to_dimension_mapping = {
+    "n_head": ("num_attention_heads",),
+    "n_kv_head": ("num_key_value_heads",), 
+    "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": ("head_dim",),
+    "attention_output": ("hidden_size",),
+    "attention_input": ("hidden_size",),
+    "query_output": ("hidden_size",),
+    "key_output": ("hidden_size",),
+    "value_output": ("hidden_size",),
+    "head_query_output": ("head_dim",),
+    "head_key_output": ("head_dim",),
+    "head_value_output": ("head_dim",),
+}
+
+"""qwen2 model with LM head"""
+qwen2_lm_type_to_module_mapping = {}
+for k, v in qwen2_type_to_module_mapping.items():
+    qwen2_lm_type_to_module_mapping[k] = (f"model.{v[0]}", ) + v[1:]
+qwen2_lm_type_to_dimension_mapping = qwen2_type_to_dimension_mapping
+
+"""qwen2 model with classifier head"""
+qwen2_classifier_type_to_module_mapping = {}
+for k, v in qwen2_type_to_module_mapping.items():
+    qwen2_classifier_type_to_module_mapping[k] = (f"model.{v[0]}", ) + v[1:]
+qwen2_classifier_type_to_dimension_mapping = qwen2_type_to_dimension_mapping
+
+def create_qwen2(
+    name="Qwen/Qwen2-7B-beta", cache_dir=None, dtype=torch.bfloat16
+):
+    """Creates a Causal LM model, config, and tokenizer from the given name and revision"""
+    from transformers import AutoConfig, AutoModelForCausalLM, AutoTokenizer
+
+    config = AutoConfig.from_pretrained(name, cache_dir=cache_dir)
+    tokenizer = AutoTokenizer.from_pretrained(name, cache_dir=cache_dir)
+    model = AutoModelForCausalLM.from_pretrained(
+        name,
+        config=config,
+        cache_dir=cache_dir,
+        torch_dtype=dtype,
+    )
+    print("loaded model")
+    return config, tokenizer, model