diff --git a/.env-example b/.env-example
index b2c620706a514ec82c3fef12209f021d926f4573..2af4cdc4ab9b8b8612cac3645a1d2c9b8f9f041d 100644
--- a/.env-example
+++ b/.env-example
@@ -1,4 +1,4 @@
-To use DeepSeek-R1, please create a HuggingFace token (for free!).
+To use HuggingFace models, please create a HuggingFace token (for free!).
Create a .env file with the following content:
HF_ACCESS_TOKEN="Your HuggingFace Token"
diff --git a/README.md b/README.md
index 5fa6054023e556185174cc7f21d0ad31067fee09..2b739c3b5311b7f9c5d80402644853f3b41ae6f0 100644
--- a/README.md
+++ b/README.md
@@ -1,18 +1,28 @@
# Named-Entity Recognition of Different Granularities
## Project Overview and Motivation
+This project explores various named entity recognition (NER) approaches, focusing on
+named entity classification (NEC).
+These methods include:
+- NLI
+- MLM (entity/class masking)
+- Word2Vec
+- LLM Prompting
+
+More about these in the report (link report here)
## Setup
+1. Run `pip install requirements.txt`
+2. If you want to use DeepSeek-R1 (via HuggingFace), follow the instructions in [`.env-example`](.env-example).
+3. Execute whatever you want to execute - the models and datasets will be downloaded automatically.
## Project Structure
+Testcases for models, datasets and individual approaches for debugging can be found in [`/tests`](tests).
-## Datasets
-
-## Models
+Models are defined in [`/models`](src/models) and are accessed via the [`common_interface`](src/common_interface.py).
-## Experiments and Results
+Datasets can be found in [`/data`](data) and are accessed via the [`data manager`](data/data_manager.py).
-## Conclusion
+Scripts for executing code on the Cluster are in [`/scripts`](scripts).
-## References
-https://aclanthology.org/P15-2048.pdf
\ No newline at end of file
+The experiments conducted during this project and some of their results are in [`/src/experiments`](src/experiments).
diff --git a/src/common_interface.py b/src/common_interface.py
index 578b666329e312b86ca43753b2b41f511e5ba663..dc6c73cd9c4d765d5bf75a801b9a109974b88dcc 100644
--- a/src/common_interface.py
+++ b/src/common_interface.py
@@ -10,7 +10,7 @@ from src.models.T5_MLM_label import classify_entity as classify_entity_t5_mlm_la
from src.models.Word2Vec import classify_entity as classify_entity_word2vec
from src.models.Word2Vec import set_label_dict as set_label_dict_word2vec
from src.experiments.NER_with_LLMs.NER_with_LLMs import find_entities as find_entities_llm
-from src.experiments.NER_with_LLMs.NER_with_LLMs import classify_entity as classify_entities_llm
+from src.experiments.NER_with_LLMs.NER_with_LLMs import classify_entity as classify_entity_llm
def classify_entity(model_name, sentence, entity, labels):
@@ -26,7 +26,7 @@ def classify_entity(model_name, sentence, entity, labels):
elif model_name == "Word2Vec":
return classify_entity_word2vec(entity, labels)
elif model_name in llms:
- return classify_entities_llm(model_name, sentence, entity, labels)
+ return classify_entity_llm(model_name, sentence, entity, labels)
else:
print(f"classify_entity not implemented for {model_name}")
@@ -42,6 +42,7 @@ def find_entities(model_name, sentence, labels):
else:
print(f"find_entities not implemented for {model_name}")
+
def set_label_dict(model_name, label_dict):
"""
NER. Sets the label dictionary required for the Word2Vec model
diff --git a/src/experiments/NER_with_LLMs/NER_with_LLMs.py b/src/experiments/NER_with_LLMs/NER_with_LLMs.py
index d1374ed3fce6dd6e59cc12ca212b82d47c3f44e0..ae0f72edfe034a86511e35a8576649dbef3636e4 100644
--- a/src/experiments/NER_with_LLMs/NER_with_LLMs.py
+++ b/src/experiments/NER_with_LLMs/NER_with_LLMs.py
@@ -18,6 +18,7 @@ Result: <answer>[('Europe', 'organization'), ('Europe', 'location'), ('German',
If there is no valid entity in the target sentence, answer "<answer>[]</answer>".
+Your Task:
"""
nli_system_prompt = """You are part of a named entity classification pipeline.
@@ -31,6 +32,7 @@ Sentence: 'Europe rejects German call to boycott British lamb.'
Target Entity: Europe
Desired Result: <answer>organization</answer>
+Your Task:
"""
diff --git a/src/models/GLiNER.py b/src/models/GLiNER.py
index 6f431d4c5dc265fcb62c5a870c9bc4fb4030cba8..35908c1a046f6c80a90a866765a016c8bfb9c42d 100644
--- a/src/models/GLiNER.py
+++ b/src/models/GLiNER.py
@@ -24,5 +24,5 @@ def classify_entity(sentence, entity, labels):
if e[0] == entity:
return e[1] # Return label
- return "Not an entity. (acc. to GLiNER)"
+ return "Target entity not found during NER."
diff --git a/src/models/llms_interface.py b/src/models/llms_interface.py
index e00690f911916dd9f43c66cb75530c833dbe37da..31ed9fc5c5373a96f477c2fcd84cc7dd3ef00a59 100644
--- a/src/models/llms_interface.py
+++ b/src/models/llms_interface.py
@@ -19,6 +19,19 @@ available_models = [
standard_system_prompt = ""
+def ensure_model_installed(model_name):
+ installed_models = ollama.list().models
+ # print(f"Installed models: {installed_models}")
+
+ installed_model_names = [m.model for m in installed_models]
+
+ if model_name not in installed_model_names:
+ print(f"Model '{model_name}' not found. Installing...")
+ ollama.pull(model_name)
+ else:
+ print(f"Model '{model_name}' is already installed.")
+
+
class LLM:
_registry = {} # Registry to store child classes
@@ -52,6 +65,7 @@ class Llama8b:
# Uses 8-Bit precision
def __init__(self,
system_prompt=standard_system_prompt):
+ ensure_model_installed("llama3.1:8b")
self.system_prompt = system_prompt
self.model_id = "ollama/llama-3.1-8b"
self.chat_history = [{"role": "system", "content": self.system_prompt}]
@@ -76,46 +90,6 @@ class Llama8b:
return self.generate_response(prompt)
-@LLM.register("Llama-3.3-70B-Instruct")
-class Llama8b:
- # Uses 8-Bit precision
- def __init__(self,
- system_prompt=standard_system_prompt):
- self.system_prompt = system_prompt
- self.client = InferenceClient(api_key=os.getenv('HF_ACCESS_TOKEN'))
- self.model_id = "meta-llama/Llama-3.3-70B-Instruct"
- self.chat_history = [{"role": "system", "content": self.system_prompt}]
- self.token_usage = []
-
- def generate_response(self, prompt):
- try:
- output = self.client.chat.completions.create(
- messages=[
- {
- "role": "user",
- "content": self.system_prompt + "\n" + prompt,
- }
- ],
- model=self.model_id,
- )
-
- token_usage = output.usage
- self.token_usage.append({
- "prompt_tokens": token_usage.prompt_tokens,
- "completion_tokens": token_usage.completion_tokens,
- "total_tokens": token_usage.total_tokens,
- })
- return output.choices[0].message.content
- except Exception as e:
- return f"Error: {e}"
-
- def __call__(self, prompt, reset=True):
- if reset:
- self.chat_history = [{"role": "system", "content": self.system_prompt}]
- # Else append new message to chat
- return self.generate_response(prompt)
-
-
@LLM.register("DeepSeek-R1-Distill-Qwen-32B")
class R1:
def __init__(self, system_prompt=standard_system_prompt):