Update README.md

README.md

@@ -1,199 +1,143 @@
 ---
 library_name: transformers
-tags: []
+license: gpl-3.0
+datasets:
+- HuggingFaceTB/Countdown-Task-GOLD
+language:
+- en
+metrics:
+- accuracy
+base_model:
+- google/gemma-3-1b-it
+pipeline_tag: text-generation
 ---
 
-# Model Card for Model ID
+# Countdown Distillation on Gemma 3 1B
 
-<!-- Provide a quick summary of what the model is/does. -->
+## Overview
 
+`google/gemma-3-1b-it` is a compact student model and a good fit for distillation. We trained it to solve Countdown-style arithmetic tasks: given a set of numbers and basic operators `(+, -, *, /)`, the model must create an equation that reaches a target value. Example:
 
+- Numbers: `[75, 80, 90, 24]`
+- Target: `61`
+- Solution: `90 - 80 + 75 - 24 = 61`
 
-## Model Details
+The student is supervised with reasoning traces, generated by `Qwen2.5-7B-Instruct`, from the Countdown [dataset](https://huggingface.co/datasets/HuggingFaceTB/Countdown-Task-GOLD) and learns to produce the final equation in `<think>` and `<answer>` format.
 
-### Model Description
+## Dataset
 
-<!-- Provide a longer summary of what this model is. -->
+The training data contains verified Countdown solutions with the following fields: `target`, `nums`, and `messages`. The final maximum sequence length is `1024` and the split is `95/5`:
 
-This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.
+- Train: `27,809` samples
+- Validation: `1,464` samples
 
-- **Developed by:** [More Information Needed]
-- **Funded by [optional]:** [More Information Needed]
-- **Shared by [optional]:** [More Information Needed]
-- **Model type:** [More Information Needed]
-- **Language(s) (NLP):** [More Information Needed]
-- **License:** [More Information Needed]
-- **Finetuned from model [optional]:** [More Information Needed]
+The token-length distribution:
 
-### Model Sources [optional]
+![output_8_0](https://cdn-uploads.huggingface.co/production/uploads/6957bafe54c6b170be4df9cb/lL9oZ0bfrX71-lBEC9-PC.png)
 
-<!-- Provide the basic links for the model. -->
+## Training
 
-- **Repository:** [More Information Needed]
-- **Paper [optional]:** [More Information Needed]
-- **Demo [optional]:** [More Information Needed]
+Distillation was performed with the following setup:
 
-## Uses
+- GPU: NVIDIA GeForce RTX 5090
+- VRAM: 31.35 GB
+- CPU: Ryzen 9 9950X
+- RAM: 64 GB
 
-<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+Training settings:
 
-### Direct Use
+- max sequence length: `1024`
+- batch size: `4`
+- gradient accumulation: `8`
+- epochs: `1`
+- learning rate: `2e-4`
+- warmup ratio: `0.1`
+- scheduler: cosine
+- optimiser: `adamw_torch`
+- LoRA rank: `16`
+- LoRA alpha: `32`
+- LoRA dropout: `0.05`
 
-<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+The best checkpoint is selected by validation loss.
 
-[More Information Needed]
+## Loss and accuracy curves
 
-### Downstream Use [optional]
+The training and validation losses show a steady downward trend and then settle near a stable plateau.
 
-<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+![output_20_0](https://cdn-uploads.huggingface.co/production/uploads/6957bafe54c6b170be4df9cb/_o5DA1SvydEIoxFMe4Puk.png)
 
-[More Information Needed]
+Also available as a logarithmic plot:
 
-### Out-of-Scope Use
+![output_21_0](https://cdn-uploads.huggingface.co/production/uploads/6957bafe54c6b170be4df9cb/iF5cntbkGrLWdo4JafYhe.png)
 
-<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+Validation accuracy gradually grows with small oscillations:
 
-[More Information Needed]
-
-## Bias, Risks, and Limitations
-
-<!-- This section is meant to convey both technical and sociotechnical limitations. -->
-
-[More Information Needed]
-
-### Recommendations
-
-<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
-
-Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
-
-## How to Get Started with the Model
-
-Use the code below to get started with the model.
-
-[More Information Needed]
-
-## Training Details
-
-### Training Data
-
-<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
-
-[More Information Needed]
-
-### Training Procedure
-
-<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
-
-#### Preprocessing [optional]
-
-[More Information Needed]
-
-
-#### Training Hyperparameters
-
-- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
-
-#### Speeds, Sizes, Times [optional]
-
-<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
-
-[More Information Needed]
+![output_22_0](https://cdn-uploads.huggingface.co/production/uploads/6957bafe54c6b170be4df9cb/rR8jUCL7lBRZ9SrPb0_LJ.png)
 
 ## Evaluation
 
-<!-- This section describes the evaluation protocols and provides the results. -->
-
-### Testing Data, Factors & Metrics
-
-#### Testing Data
-
-<!-- This should link to a Dataset Card if possible. -->
-
-[More Information Needed]
-
-#### Factors
-
-<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
-
-[More Information Needed]
-
-#### Metrics
-
-<!-- These are the evaluation metrics being used, ideally with a description of why. -->
-
-[More Information Needed]
-
-### Results
-
-[More Information Needed]
-
-#### Summary
-
-
-
-## Model Examination [optional]
-
-<!-- Relevant interpretability work for the model goes here -->
-
-[More Information Needed]
-
-## Environmental Impact
-
-<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
-
-Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
-
-- **Hardware Type:** [More Information Needed]
-- **Hours used:** [More Information Needed]
-- **Cloud Provider:** [More Information Needed]
-- **Compute Region:** [More Information Needed]
-- **Carbon Emitted:** [More Information Needed]
-
-## Technical Specifications [optional]
-
-### Model Architecture and Objective
-
-[More Information Needed]
-
-### Compute Infrastructure
-
-[More Information Needed]
-
-#### Hardware
-
-[More Information Needed]
-
-#### Software
-
-[More Information Needed]
+Validation was run on the first `1,000` examples of the validation split with batch size `200`.
 
-## Citation [optional]
+The final result is:
 
-<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+- Validation accuracy: `0.8200` (`820/1000`)
 
-**BibTeX:**
+## Inference
 
-[More Information Needed]
+Use these two cells for inference.
 
-**APA:**
+```python
+import torch
+from transformers import AutoTokenizer, AutoModelForCausalLM
+from peft import PeftModel
 
-[More Information Needed]
+base_model_id = "google/gemma-3-1b-it"
+adapter_id = "pymlex/gemma3-1b-countdown"
 
-## Glossary [optional]
+tokenizer = AutoTokenizer.from_pretrained(base_model_id, trust_remote_code=True)
+if tokenizer.pad_token is None:
+    tokenizer.pad_token = tokenizer.eos_token
+tokenizer.padding_side = "left"
 
-<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+base_model = AutoModelForCausalLM.from_pretrained(
+    base_model_id,
+    device_map="auto",
+    torch_dtype=torch.bfloat16 if torch.cuda.is_available() and torch.cuda.is_bf16_supported() else torch.float16,
+    trust_remote_code=True,
+)
 
-[More Information Needed]
+model = PeftModel.from_pretrained(base_model, adapter_id)
+model.eval()
+````
 
-## More Information [optional]
+```python
+def generate_continuation(model, tokenizer, prompt, max_new_tokens=850):
+    inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
+    prompt_len = inputs.input_ids.shape[1]
 
-[More Information Needed]
+    outputs = model.generate(
+        **inputs,
+        max_new_tokens=max_new_tokens,
+        temperature=0.7,
+        top_p=0.95,
+        do_sample=True,
+        repetition_penalty=1.05,
+        eos_token_id=tokenizer.eos_token_id,
+        pad_token_id=tokenizer.pad_token_id,
+    )
 
-## Model Card Authors [optional]
+    decoded = tokenizer.decode(outputs[0][prompt_len:], skip_special_tokens=True)
+    return decoded.strip()
 
-[More Information Needed]
 
-## Model Card Contact
+sample_prompt = (
+    "Using the numbers [78, 46, 93], create an equation that equals 61. "
+    "You can use basic arithmetic operations (+, -, *, /) and each number can only be used once."
+)
 
-[More Information Needed]
+output = generate_continuation(model, tokenizer, sample_prompt, max_new_tokens=850)
+print("Prompt:")
+print(sample_prompt)
+print("\nGenerated continuation:")
+print(output)
+```