Instructions to use summerMC/TRM-textV2 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use summerMC/TRM-textV2 with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="summerMC/TRM-textV2", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("summerMC/TRM-textV2", trust_remote_code=True, device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use summerMC/TRM-textV2 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "summerMC/TRM-textV2"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "summerMC/TRM-textV2",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/summerMC/TRM-textV2

SGLang

How to use summerMC/TRM-textV2 with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "summerMC/TRM-textV2" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "summerMC/TRM-textV2",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "summerMC/TRM-textV2" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "summerMC/TRM-textV2",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use summerMC/TRM-textV2 with Docker Model Runner:
```
docker model run hf.co/summerMC/TRM-textV2
```

summerMC commited on Jun 7

Commit

dc244cf

verified ·

1 Parent(s): 29ecdb7

v36: Enhanced README and permanent chat_template integration

Browse files

Files changed (3) hide show

README.md +23 -8
chat_template.jinja +1 -0
tokenizer_config.json +1 -1

README.md CHANGED Viewed

@@ -6,21 +6,36 @@ tags:
 - text-generation
 - trm-text
 - ism
 ---
-# TRM-textV2: Recurrent Shared Transformer with ISM
-This model is a Recurrent Shared Transformer trained with the Inverse Square Mask (ISM) logic. It uses a single Transformer block repeated multiple times (recurrence_steps=4) to simulate depth while maintaining a lower parameter count.
-## Key Features
-- **Architecture**: Shared Recurrent Transformer Block (v34).
-- **Inference**: Supports ISM-based prefix-answer masking.
-- **Training**: TinyStories & FineWeb optimized.
-## Usage
 ```python
 from transformers import AutoModelForCausalLM, AutoTokenizer
 model = AutoModelForCausalLM.from_pretrained('summerMC/TRM-textV2', trust_remote_code=True)
-tokenizer = AutoTokenizer.from_pretrained('summerMC/TRM-textV2')
 ```

 - text-generation
 - trm-text
 - ism
+- recurrent-transformer
+- tiny-stories
+library_name: transformers
+metrics:
+- accuracy
 ---
+# 🤖 TRM-textV2: Recurrent Shared Transformer
+TRM-textV2 is a high-efficiency language model featuring a **Shared Recurrent Transformer** architecture enhanced with **Inverse Square Mask (ISM)** logic.
+## 🌟 Model Highlights
+- **Efficient Depth**: Simulates a deep network by repeating a single Transformer block (recurrence_steps=4).
+- **ISM Integration**: Advanced prefix-answer masking for superior long-range dependency handling.
+- **Optimized for Stability**: Trained with specific residual scaling and gate initialization to prevent loss plateaus.
+## 🚀 Quick Start
 ```python
 from transformers import AutoModelForCausalLM, AutoTokenizer
+tokenizer = AutoTokenizer.from_pretrained('summerMC/TRM-textV2', trust_remote_code=True)
 model = AutoModelForCausalLM.from_pretrained('summerMC/TRM-textV2', trust_remote_code=True)
+# Standard Chat Template use
+messages = [{'role': 'user', 'content': 'Once upon a time, a small robot'}]
+inputs = tokenizer.apply_chat_template(messages, add_generation_prompt=True, return_tensors='pt')
 ```
+## 📊 Training Details
+- **Dataset**: TinyStories & FineWeb-Edu
+- **Architecture**: 45M parameters (Effective depth equivalent to larger models)
+- **License**: MIT

chat_template.jinja ADDED Viewed

	@@ -0,0 +1 @@


1	+ {% for message in messages %}{{'<\|' + message['role'] + '\|>\n' + message['content'] + '<\|end\|>\n'}}{% endfor %}

tokenizer_config.json CHANGED Viewed

@@ -4,7 +4,7 @@
   "bos_token": "<|endoftext|>",
   "eos_token": "<|endoftext|>",
   "errors": "replace",
-  "is_local": false,
   "local_files_only": false,
   "model_max_length": 1024,
   "pad_token": null,

   "bos_token": "<|endoftext|>",
   "eos_token": "<|endoftext|>",
   "errors": "replace",
+  "is_local": true,
   "local_files_only": false,
   "model_max_length": 1024,
   "pad_token": null,