Training in progress, step 90

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+tokenizer.json filter=lfs diff=lfs merge=lfs -text

README.md

@@ -1,105 +1,58 @@
-# Tiny Coder Prompt Completion 0.5B
+---
+base_model: Qwen/Qwen2.5-Coder-0.5B-Instruct
+library_name: transformers
+model_name: tiny-coder-prompt-completion-0.5B
+tags:
+- generated_from_trainer
+- trl
+- sft
+licence: license
+---
 
-A tiny model (~494M parameters, ~942MB weights) for **prompt autocomplete** in agentic coding editors like Claude Code, Codex CLI, aider, etc.
+# Model Card for tiny-coder-prompt-completion-0.5B
 
-## What it does
+This model is a fine-tuned version of [Qwen/Qwen2.5-Coder-0.5B-Instruct](https://huggingface.co/Qwen/Qwen2.5-Coder-0.5B-Instruct).
+It has been trained using [TRL](https://github.com/huggingface/trl).
 
-When you're typing a prompt to an AI coding agent, this model suggests how to **complete your thought**. For example:
-
-- `fix the bug in src/utils` → `.py where the auth token isn't refreshing`
-- `refactor the database` → `connection logic to use connection pooling`
-- `implement caching for the` → `user profile endpoint using Redis with a 5-minute TTL`
-
-## Base Model
-
-This is a fine-tuned version of [Qwen2.5-Coder-0.5B-Instruct](https://huggingface.co/Qwen/Qwen2.5-Coder-0.5B-Instruct), which was purpose-built for code completion.
-
-## On-Device Requirements
-
-- **RAM**: ~1GB for fp16 inference (easily fits in 16GB MacBooks)
-- **Disk**: ~1GB for model weights
-- **CPU/GPU**: Works great on CPU (M1/M2/M3 MacBooks), even better with GPU
-
-## Quick Start
-
-### Python (transformers)
+## Quick start
 
 ```python
-from transformers import AutoTokenizer, AutoModelForCausalLM
-import torch
-
-model_id = "finnvoorhees/tiny-coder-prompt-completion-0.5B"
-tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
-model = AutoModelForCausalLM.from_pretrained(
-    model_id,
-    trust_remote_code=True,
-    torch_dtype=torch.bfloat16,
-    device_map="auto",
-)
-
-def complete_prompt(prompt: str) -> str:
-    messages = [
-        {"role": "system", "content": "You are a helpful coding assistant. Complete the user's partial prompt concisely."},
-        {"role": "user", "content": prompt},
-    ]
-    inputs = tokenizer.apply_chat_template(
-        messages,
-        tokenize=True,
-        return_tensors="pt",
-        return_dict=True,
-        add_generation_prompt=True,
-    ).to(model.device)
-    
-    outputs = model.generate(
-        **inputs,
-        max_new_tokens=32,
-        temperature=0.3,
-        top_p=0.9,
-        do_sample=True,
-    )
-    
-    new_tokens = outputs[0][inputs["input_ids"].shape[1]:]
-    return tokenizer.decode(new_tokens, skip_special_tokens=True).strip()
+from transformers import pipeline
 
-print(complete_prompt("refactor the database"))
-# → "connection logic to use connection pooling instead of creating new connections each time"
+question = "If you had a time machine, but could only go to the past or the future once and never return, which would you choose and why?"
+generator = pipeline("text-generation", model="finnvoorhees/tiny-coder-prompt-completion-0.5B", device="cuda")
+output = generator([{"role": "user", "content": question}], max_new_tokens=128, return_full_text=False)[0]
+print(output["generated_text"])
 ```
 
-### Even smaller: GGUF / llama.cpp
+## Training procedure
 
-For ultra-fast CPU inference, convert to GGUF (Q4_K_M quantization → ~300MB):
+ 
 
-```bash
-# Install llama.cpp
-brew install llama.cpp
 
-# Convert to GGUF (or download from the GGUF tag on this repo)
-python convert_hf_to_gguf.py finnvoorhees/tiny-coder-prompt-completion-0.5B --outfile tiny-coder-0.5b-Q4_K_M.gguf
 
-# Run inference
-llama-cli -m tiny-coder-0.5b-Q4_K_M.gguf -p "fix the bug in src/utils"
-```
+This model was trained with SFT.
 
-## Integration with Agentic Editors
+### Framework versions
 
-### Claude Code
-Create a shell script that calls this model and pipe completions into your prompt:
+- TRL: 1.2.0
+- Transformers: 5.6.2
+- Pytorch: 2.11.0
+- Datasets: 4.8.4
+- Tokenizers: 0.22.2
 
-```bash
-# ~/.claude-complete.sh
-python -c "from transformers import pipeline; print(pipeline('text-generation', model='finnvoorhees/tiny-coder-prompt-completion-0.5B')('$1'))"
-```
+## Citations
 
-### Codex CLI / aider
-Set up a local API endpoint using `transformers` or `llama.cpp` server mode:
 
-```bash
-# llama.cpp server (fast!)
-llama-server -m tiny-coder-0.5b-Q4_K_M.gguf --port 8080
 
-# Then use http://localhost:8080/v1/completions in your editor
-```
-
-## License
-
-Apache-2.0 (same as the base Qwen2.5-Coder-0.5B-Instruct model)
+Cite TRL as:
+    
+```bibtex
+@software{vonwerra2020trl,
+  title   = {{TRL: Transformers Reinforcement Learning}},
+  author  = {von Werra, Leandro and Belkada, Younes and Tunstall, Lewis and Beeching, Edward and Thrush, Tristan and Lambert, Nathan and Huang, Shengyi and Rasul, Kashif and Gallouédec, Quentin},
+  license = {Apache-2.0},
+  url     = {https://github.com/huggingface/trl},
+  year    = {2020}
+}
+```

chat_template.jinja

@@ -0,0 +1,54 @@
+{%- if tools %}
+    {{- '<|im_start|>system\n' }}
+    {%- if messages[0]['role'] == 'system' %}
+        {{- messages[0]['content'] }}
+    {%- else %}
+        {{- 'You are Qwen, created by Alibaba Cloud. You are a helpful assistant.' }}
+    {%- endif %}
+    {{- "\n\n# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
+{%- else %}
+    {%- if messages[0]['role'] == 'system' %}
+        {{- '<|im_start|>system\n' + messages[0]['content'] + '<|im_end|>\n' }}
+    {%- else %}
+        {{- '<|im_start|>system\nYou are Qwen, created by Alibaba Cloud. You are a helpful assistant.<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- for message in messages %}
+    {%- if (message.role == "user") or (message.role == "system" and not loop.first) or (message.role == "assistant" and not message.tool_calls) %}
+        {{- '<|im_start|>' + message.role + '\n' + message.content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {{- '<|im_start|>' + message.role }}
+        {%- if message.content %}
+            {{- '\n' + message.content }}
+        {%- endif %}
+        {%- for tool_call in message.tool_calls %}
+            {%- if tool_call.function is defined %}
+                {%- set tool_call = tool_call.function %}
+            {%- endif %}
+            {{- '\n<tool_call>\n{"name": "' }}
+            {{- tool_call.name }}
+            {{- '", "arguments": ' }}
+            {{- tool_call.arguments | tojson }}
+            {{- '}\n</tool_call>' }}
+        {%- endfor %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if (loop.index0 == 0) or (messages[loop.index0 - 1].role != "tool") %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- message.content }}
+        {{- '\n</tool_response>' }}
+        {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+{%- endif %}

config.json

@@ -0,0 +1,57 @@
+{
+  "architectures": [
+    "Qwen2ForCausalLM"
+  ],
+  "attention_dropout": 0.0,
+  "bos_token_id": null,
+  "dtype": "bfloat16",
+  "eos_token_id": 151645,
+  "hidden_act": "silu",
+  "hidden_size": 896,
+  "initializer_range": 0.02,
+  "intermediate_size": 4864,
+  "layer_types": [
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention"
+  ],
+  "max_position_embeddings": 32768,
+  "max_window_layers": 24,
+  "model_type": "qwen2",
+  "num_attention_heads": 14,
+  "num_hidden_layers": 24,
+  "num_key_value_heads": 2,
+  "pad_token_id": 151643,
+  "rms_norm_eps": 1e-06,
+  "rope_parameters": {
+    "rope_theta": 1000000.0,
+    "rope_type": "default"
+  },
+  "sliding_window": null,
+  "tie_word_embeddings": true,
+  "transformers_version": "5.6.2",
+  "use_cache": false,
+  "use_sliding_window": false,
+  "vocab_size": 151936
+}

generation_config.json

@@ -0,0 +1,13 @@
+{
+  "do_sample": true,
+  "eos_token_id": [
+    151645,
+    151643
+  ],
+  "pad_token_id": 151643,
+  "repetition_penalty": 1.05,
+  "temperature": 0.7,
+  "top_k": 20,
+  "top_p": 0.8,
+  "transformers_version": "5.6.2"
+}

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:81513daf667ea3ce13c2a16e0b4ed2631ae989f74efe875217d6422c66f0f20e
+size 988097824

tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3fd169731d2cbde95e10bf356d66d5997fd885dd8dbb6fb4684da3f23b2585d8
+size 11421892

tokenizer_config.json

@@ -0,0 +1,30 @@
+{
+  "add_prefix_space": false,
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|im_end|>",
+  "errors": "replace",
+  "extra_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>",
+    "<|object_ref_start|>",
+    "<|object_ref_end|>",
+    "<|box_start|>",
+    "<|box_end|>",
+    "<|quad_start|>",
+    "<|quad_end|>",
+    "<|vision_start|>",
+    "<|vision_end|>",
+    "<|vision_pad|>",
+    "<|image_pad|>",
+    "<|video_pad|>"
+  ],
+  "is_local": false,
+  "local_files_only": false,
+  "model_max_length": 32768,
+  "pad_token": "<|endoftext|>",
+  "split_special_tokens": false,
+  "tokenizer_class": "Qwen2Tokenizer",
+  "unk_token": null
+}

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:af1e287adb66fc7b18c7e1c19396bba4e849d9c353d7e3bf759308521994fa6d
+size 5841