tool-bert

README.md

@@ -15,64 +15,49 @@ should probably proofread and complete it, then remove this comment. -->
 
 # tool-bert
 
-This model is a fine-tuned version of [google-bert/bert-base-uncased](https://huggingface.co/google-bert/bert-base-uncased).
+This model is a fine-tuned version of [google-bert/bert-base-uncased](https://huggingface.co/google-bert/bert-base-uncased) on an unknown dataset.
+It achieves the following results on the evaluation set:
+- Loss: 0.0141
+- Accuracy: 1.0
 
-It uses a custom made dataset of sample user instructions, which are classified to a number of possible local assistant function calling endpoints. 
+## Model description
 
-For example, given an input query, tool-bert returns a prediction as to what tool to use to augment a downstream LLM generated output with. 
+More information needed
 
-More information on these tools to follow, but example tools are "play music", "check the weather", "get the news", "take a photo", or use no tool. 
+## Intended uses & limitations
 
-Basically, this model is meant to be a means of allowing very small LLMs (i.e. 8B and below) to use function calling. 
+More information needed
 
-All limitations and biases are inherited from the parent model. 
+## Training and evaluation data
 
-### Example Usage
+More information needed
 
-```python
-from transformers import AutoTokenizer
-from transformers import AutoModelForSequenceClassification
+## Training procedure
 
-key_tools = ['take_picture', 'no_tool_needed',
-             'check_news', 'check_weather', 'play_spotify']
+### Training hyperparameters
 
+The following hyperparameters were used during training:
+- learning_rate: 5e-05
+- train_batch_size: 8
+- eval_batch_size: 8
+- seed: 42
+- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
+- lr_scheduler_type: linear
+- num_epochs: 4
 
-def get_id2tool_name(id, key_tools):
-    return key_tools[id]
+### Training results
 
+| Training Loss | Epoch | Step | Validation Loss | Accuracy |
+|:-------------:|:-----:|:----:|:---------------:|:--------:|
+| No log        | 1.0   | 32   | 0.5841          | 0.824    |
+| No log        | 2.0   | 64   | 0.0964          | 1.0      |
+| No log        | 3.0   | 96   | 0.0206          | 1.0      |
+| No log        | 4.0   | 128  | 0.0141          | 1.0      |
 
-def remove_any_non_alphanumeric_characters(text):
-    return ''.join(e for e in text if e.isalnum() or e.isspace())
 
-
-def load_model():
-    tokenizer = AutoTokenizer.from_pretrained("google-bert/bert-base-uncased")
-    model = AutoModelForSequenceClassification.from_pretrained(
-        "nkasmanoff/tool-bert")
-
-    model.eval()
-    return model, tokenizer
-
-
-def predict_tool(question, model, tokenizer):
-    question = remove_any_non_alphanumeric_characters(question)
-    inputs = tokenizer(question, return_tensors="pt")
-
-    outputs = model(**inputs)
-
-    logits = outputs.logits
-    return get_id2tool_name(logits.argmax().item(), key_tools)
-
-model, tokenizer = load_model()
-
-question = "What's the weather outside?"
-
-predict_tool(question, model, tokenizer)
-> check_weather
-```
 ### Framework versions
 
 - Transformers 4.41.1
 - Pytorch 2.3.0
 - Datasets 2.19.1
-- Tokenizers 0.19.1
+- Tokenizers 0.19.1

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b162d53eb5bb90c42878d8b54e388c4117f41af0bfe39c51d5ca5635f98617e9
+oid sha256:67eddbca69bbac71d63c5b245f3b9725fcc35e2967d421735515b7ded4f7ded6
 size 437967876

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cd0aeb6003fce2a4fec9c124df29b1584ebe21671f747746295f69d20a338b19
+oid sha256:4caa93fffa53fd64ebdf6c5724a7f4c15dc5a635ad77f5a77997944eaf19bf6f
 size 5048