templates

src/text_generation/adapters/foundation_models/apple_openelm_foundation_model.py

@@ -6,12 +6,12 @@ from src.text_generation.common.model_id import ModelId
 
 
 class AppleOpenELMFoundationModel(BaseFoundationModel):
-    """Apple OpenELM 270M implementation"""
+    """apple/OpenELM-3B-Instruct implementation"""
 
     def __init__(self, config: AppleOpenELMConfig = AppleOpenELMConfig()):
         self.config = config
         super().__init__()
-        self.MODEL_ID = ModelId.APPLE_OPENELM_270M_INSTRUCT.value
+        self.MODEL_ID = ModelId.APPLE_OPENELM_3B_INSTRUCT.value
 
     def _load_model(self) -> None:
         self.tokenizer = AutoTokenizer.from_pretrained(

src/text_generation/adapters/foundation_models/base/base_model_pipeline.py

@@ -1,5 +1,5 @@
 from transformers import pipeline
-from langchain.llms import HuggingFacePipeline
+from langchain_huggingface import HuggingFacePipeline
 from typing import Dict, Any, List
 from abc import ABC, abstractmethod
 

src/text_generation/adapters/foundation_models/config/meta_llama_config.py

@@ -4,7 +4,7 @@ from src.text_generation.adapters.foundation_models.base.base_model_config impor
 
 
 @dataclass
-class MetaTinyLlamaConfig(BaseModelConfig):
-    """TinyLlama-specific configuration"""
+class MetaLlamaConfig(BaseModelConfig):
+    """meta-llama/Llama-3.2-3B-Instruct configuration"""
     use_flash_attention: bool = False
     rope_scaling: Optional[Dict[str, Any]] = None

src/text_generation/adapters/foundation_models/factories/foundation_model_factory.py

@@ -5,10 +5,10 @@ from langchain_huggingface import HuggingFacePipeline
 from src.text_generation.adapters.foundation_models.apple_openelm_foundation_model import AppleOpenELMFoundationModel
 from src.text_generation.adapters.foundation_models.base.base_foundation_model import BaseFoundationModel
 from src.text_generation.adapters.foundation_models.base.base_model_config import BaseModelConfig
-from src.text_generation.adapters.foundation_models.meta_tinyllama_foundation_model import MetaTinyLlamaFoundationModel
+from src.text_generation.adapters.foundation_models.meta_llama_foundation_model import MetaLlamaFoundationModel
 from src.text_generation.adapters.foundation_models.microsoft_phi3_foundation_model import MicrosoftPhi3FoundationModel
 from src.text_generation.adapters.foundation_models.pipelines.apple_openelm_pipeline import AppleOpenELMPipeline
-from src.text_generation.adapters.foundation_models.pipelines.meta_tinyllama_pipeline import MetaTinyLlamaPipeline
+from src.text_generation.adapters.foundation_models.pipelines.meta_llama_pipeline import MetaLlamaPipeline
 from src.text_generation.adapters.foundation_models.pipelines.microsoft_phi3mini_pipeline import MicrosoftPhi3MiniPipeline
 from src.text_generation.common.model_id import ModelId
 
@@ -22,8 +22,8 @@ class FoundationModelFactory:
             config = BaseModelConfig()
 
         model_map = {
-            ModelId.APPLE_OPENELM_270M_INSTRUCT.value: AppleOpenELMFoundationModel,
-            ModelId.META_TINYLLAMA_1_1B_CHAT.value: MetaTinyLlamaFoundationModel,
+            ModelId.APPLE_OPENELM_3B_INSTRUCT.value: AppleOpenELMFoundationModel,
+            ModelId.META_LLAMA_3_2_3B_INSTRUCT.value: MetaLlamaFoundationModel,
             ModelId.MICROSOFT_PHI_3_MINI4K_INSTRUCT.value: MicrosoftPhi3FoundationModel
         }
 
@@ -37,8 +37,8 @@ class FoundationModelFactory:
         """Factory function to create the appropriate pipeline based on model name"""
         
         pipeline_classes = {
-            ModelId.APPLE_OPENELM_270M_INSTRUCT.value: AppleOpenELMPipeline,
-            ModelId.META_TINYLLAMA_1_1B_CHAT.value: MetaTinyLlamaPipeline,
+            ModelId.APPLE_OPENELM_3B_INSTRUCT.value: AppleOpenELMPipeline,
+            ModelId.META_LLAMA_3_2_3B_INSTRUCT.value: MetaLlamaPipeline,
             ModelId.MICROSOFT_PHI_3_MINI4K_INSTRUCT.value: MicrosoftPhi3MiniPipeline
         }
         

src/text_generation/adapters/foundation_models/factory/foundation_model_factory.py

@@ -3,7 +3,7 @@ from typing import Optional
 from src.text_generation.adapters.foundation_models.apple_openelm_foundation_model import AppleOpenELMFoundationModel
 from src.text_generation.adapters.foundation_models.base.base_foundation_model import BaseFoundationModel
 from src.text_generation.adapters.foundation_models.base.base_model_config import BaseModelConfig
-from src.text_generation.adapters.foundation_models.meta_tinyllama_foundation_model import MetaTinyLlamaFoundationModel
+from src.text_generation.adapters.foundation_models.meta_llama_foundation_model import MetaLlamaFoundationModel
 from src.text_generation.adapters.foundation_models.microsoft_phi3_foundation_model import MicrosoftPhi3FoundationModel
 from src.text_generation.common.model_id import ModelId
 
@@ -17,8 +17,8 @@ class FoundationModelFactory:
             config = BaseModelConfig()
 
         model_map = {
-            ModelId.APPLE_OPENELM_270M_INSTRUCT.value: AppleOpenELMFoundationModel,
-            ModelId.META_TINYLLAMA_1_1B_CHAT.value: MetaTinyLlamaFoundationModel,
+            ModelId.APPLE_OPENELM_3B_INSTRUCT.value: AppleOpenELMFoundationModel,
+            ModelId.META_LLAMA_3_2_3B_INSTRUCT.value: MetaLlamaFoundationModel,
             ModelId.MICROSOFT_PHI_3_MINI4K_INSTRUCT.value: MicrosoftPhi3FoundationModel
         }
 

src/text_generation/adapters/foundation_models/meta_llama_foundation_model.py

@@ -1,4 +1,4 @@
-from src.text_generation.adapters.foundation_models.config.meta_tinyllama_config import MetaTinyLlamaConfig
+from src.text_generation.adapters.foundation_models.config.meta_llama_config import MetaLlamaConfig
 from src.text_generation.adapters.foundation_models.base.base_foundation_model import BaseFoundationModel
 
 
@@ -8,13 +8,13 @@ from transformers import AutoModelForCausalLM, AutoTokenizer
 from src.text_generation.common.model_id import ModelId
 
 
-class MetaTinyLlamaFoundationModel(BaseFoundationModel):
-    """Meta TinyLlama 1.1B implementation"""
+class MetaLlamaFoundationModel(BaseFoundationModel):
+    """meta-llama/Llama-3.2-3B-Instruct implementation"""
 
-    def __init__(self, config: MetaTinyLlamaConfig = MetaTinyLlamaConfig()):
+    def __init__(self, config: MetaLlamaConfig = MetaLlamaConfig()):
         self.config = config
         super().__init__()
-        self.MODEL_ID = ModelId.META_TINYLLAMA_1_1B_CHAT.value
+        self.MODEL_ID = ModelId.META_LLAMA_3_2_3B_INSTRUCT.value
 
     def _load_model(self) -> None:
         self.tokenizer = AutoTokenizer.from_pretrained(

src/text_generation/adapters/foundation_models/pipelines/meta_llama_pipeline.py

@@ -4,11 +4,11 @@ from src.text_generation.adapters.foundation_models.base.base_model_pipeline imp
 from typing import Any, Dict, List
 
 
-class MetaTinyLlamaPipeline(BaseModelPipeline):
+class MetaLlamaPipeline(BaseModelPipeline):
     def get_model_specific_config(self) -> Dict[str, Any]:
         return {
             "max_new_tokens": 512,
-            # TinyLlama might need slightly different settings
+            # meta-llama/Llama-3.2-3B-Instruct might need slightly different settings
             "top_p": 0.9,  # Add nucleus sampling for better diversity
         }
 

src/text_generation/adapters/prompt_template_repository.py

@@ -21,5 +21,5 @@ class PromptTemplateRepository(AbstractPromptTemplateRepository):
             return None
     
     def add(self, id: str, prompt_template: PromptTemplate) -> None:
-        if self.get(id) == None:
-            prompt_template.save(self._create_path_from_id(id))
+        print(f'Saving template: {id}')
+        prompt_template.save(self._create_path_from_id(id))

src/text_generation/common/constants.py

@@ -6,16 +6,13 @@ class Constants:
     PHI_3_USER_START_TOKEN =      "<|user|>"
 
     # OpenELM tokens
-    OPENELM_SYSTEM_START_TOKEN =  "[INST]"
-    OPENELM_USER_START_TOKEN =    "[INST]"
-    OPENELM_ASSISTANT_TOKEN =     "[/INST]"
     OPENELM_END_TOKEN =           "</s>"
 
-    # TinyLlama tokens
-    TINYLLAMA_SYSTEM_START_TOKEN = "<|system|>"
-    TINYLLAMA_USER_START_TOKEN =   "<|user|>"
-    TINYLLAMA_ASSISTANT_TOKEN =    "<|assistant|>"
-    TINYLLAMA_END_TOKEN =          "</s>"
+    # meta-llama/Llama-3.2-3B-Instruct tokens
+    LLAMA_SYSTEM_START_TOKEN = "### SYSTEM:\n"
+    LLAMA_USER_START_TOKEN =   "### HUMAN:\n"
+    LLAMA_ASSISTANT_TOKEN =    "### ASSISTANT:\n"
+    LLAMA_END_TOKEN =          ""
 
     INPUT_VARIABLE_TOKEN =        "input"
     ALT_COMPLETION_TEXT =         "I can't answer that."
@@ -28,16 +25,16 @@ class Constants:
         PHI_3_MINI_4K_INSTRUCT__04_FEW_SHOT_RAG_PLUS_COT =      "phi-3-mini-4k-instruct.04-few-shot-rag-plus-cot"
         PHI_3_MINI_4K_INSTRUCT__05_REFLEXION =                  "phi-3-mini-4k-instruct.05-reflexion"
 
-        # OpenELM templates
-        OPENELM_270M_INSTRUCT__01_BASIC =                      "openelm-270m-instruct.01-basic"
-        OPENELM_270M_INSTRUCT__02_ZERO_SHOT_CHAIN_OF_THOUGHT = "openelm-270m-instruct.02-zero-shot-cot"
-        OPENELM_270M_INSTRUCT__03_FEW_SHOT_EXAMPLES =          "openelm-270m-instruct.03-few-shot"
-        OPENELM_270M_INSTRUCT__04_FEW_SHOT_RAG_PLUS_COT =      "openelm-270m-instruct.04-few-shot-rag-plus-cot"
-        OPENELM_270M_INSTRUCT__05_REFLEXION =                  "openelm-270m-instruct.05-reflexion"
+        # OpenELM templates: apple/OpenELM-3B-Instruct
+        OPENELM_3B_INSTRUCT__01_BASIC =                      "openelm-3b-instruct.01-basic"
+        OPENELM_3B_INSTRUCT__02_ZERO_SHOT_CHAIN_OF_THOUGHT = "openelm-3b-instruct.02-zero-shot-cot"
+        OPENELM_3B_INSTRUCT__03_FEW_SHOT_EXAMPLES =          "openelm-3b-instruct.03-few-shot"
+        OPENELM_3B_INSTRUCT__04_FEW_SHOT_RAG_PLUS_COT =      "openelm-3b-instruct.04-few-shot-rag-plus-cot"
+        OPENELM_3B_INSTRUCT__05_REFLEXION =                  "openelm-3b-instruct.05-reflexion"
 
-        # TinyLlama templates
-        TINYLLAMA_1_1B_CHAT__01_BASIC =                      "tinyllama-1.1b-chat.01-basic"
-        TINYLLAMA_1_1B_CHAT__02_ZERO_SHOT_CHAIN_OF_THOUGHT = "tinyllama-1.1b-chat.02-zero-shot-cot"
-        TINYLLAMA_1_1B_CHAT__03_FEW_SHOT_EXAMPLES =          "tinyllama-1.1b-chat.03-few-shot"
-        TINYLLAMA_1_1B_CHAT__04_FEW_SHOT_RAG_PLUS_COT =      "tinyllama-1.1b-chat.04-few-shot-rag-plus-cot"
-        TINYLLAMA_1_1B_CHAT__05_REFLEXION =                  "tinyllama-1.1b-chat.05-reflexion"
+        # meta-llama/Llama-3.2-3B-Instruct templates
+        LLAMA_1_1B_CHAT__01_BASIC =                      "llama-3.2-3b-instruct.01-basic"
+        LLAMA_1_1B_CHAT__02_ZERO_SHOT_CHAIN_OF_THOUGHT = "llama-3.2-3b-instruct.02-zero-shot-cot"
+        LLAMA_1_1B_CHAT__03_FEW_SHOT_EXAMPLES =          "llama-3.2-3b-instruct.03-few-shot"
+        LLAMA_1_1B_CHAT__04_FEW_SHOT_RAG_PLUS_COT =      "llama-3.2-3b-instruct.04-few-shot-rag-plus-cot"
+        LLAMA_1_1B_CHAT__05_REFLEXION =                  "llama-3.2-3b-instruct.05-reflexion"

src/text_generation/common/model_id.py

@@ -2,6 +2,6 @@ from enum import Enum
 
 
 class ModelId(Enum):
-    APPLE_OPENELM_270M_INSTRUCT = "apple/openelm-270m-instruct"
-    META_TINYLLAMA_1_1B_CHAT = "TinyLlama/TinyLlama-1.1B-Chat-v1.0"
+    APPLE_OPENELM_3B_INSTRUCT = "apple/OpenELM-3B-Instruct"
+    META_LLAMA_3_2_3B_INSTRUCT = "meta-llama/Llama-3.2-3B-Instruct"
     MICROSOFT_PHI_3_MINI4K_INSTRUCT = "microsoft/Phi-3-mini-4k-instruct-onnx"