Instructions to use Girinath11/recursive-language-model-198m with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Girinath11/recursive-language-model-198m with Transformers:

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

pipe = pipeline("text-generation", model="Girinath11/recursive-language-model-198m", trust_remote_code=True)

# Load model directly
from transformers import AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("Girinath11/recursive-language-model-198m", trust_remote_code=True, device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use Girinath11/recursive-language-model-198m with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Girinath11/recursive-language-model-198m"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Girinath11/recursive-language-model-198m",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/Girinath11/recursive-language-model-198m

SGLang

How to use Girinath11/recursive-language-model-198m 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 "Girinath11/recursive-language-model-198m" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Girinath11/recursive-language-model-198m",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

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 "Girinath11/recursive-language-model-198m" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Girinath11/recursive-language-model-198m",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use Girinath11/recursive-language-model-198m with Docker Model Runner:
```
docker model run hf.co/Girinath11/recursive-language-model-198m
```

Girinath11 commited on Mar 31

Commit

bc543bc

verified ·

1 Parent(s): c4ab5ca

Update README.md

Browse files

Files changed (1) hide show

README.md +28 -28

README.md CHANGED Viewed

@@ -27,7 +27,7 @@ pipeline_tag: text-generation
 ---
-## 🔥 Novel Architecture: Mixture of Recursion (MoR)
 Traditional transformers apply the same depth to every input — simple and complex sentences both
 go through all N layers. This wastes compute on easy inputs and under-processes hard ones.
@@ -101,16 +101,16 @@ GPT-2 Medium evaluated on identical samples for fair comparison:
 ### What the Results Actually Show
 ```
-✅ Architecture works   — training is stable, loss decreases
-✅ Router works         — 0 NaN batches, routing signals valid
-✅ Novel contribution   — perplexity-based self-supervised routing
-⚠️  Generalization gap  — val PPL 15.37 vs test PPL 43.21
-⚠️  More data needed    — 150K samples is small for this model size
 ```
 ---
-## 🚀 Quick Start
 ```bash
 pip install transformers torch
@@ -166,7 +166,7 @@ print(chat("Explain neural networks simply"))
 ---
-## 📚 Training Details
 ### Dataset — 150K Conversational Samples
@@ -203,7 +203,7 @@ Loss:
 ---
-## 💡 Technical Innovation
 ### Self-Supervised Perplexity Routing
@@ -230,21 +230,21 @@ This model is my **first from-scratch LLM** — I learned a lot:
 ```
 What worked:
-  ✅ NaN-safe attention masking (-1e4)
-  ✅ Self-supervised routing signal
-  ✅ Stable FP16 training (0 NaN batches)
-  ✅ Perplexity drops meaningfully across epochs
 What needs improvement:
-  ⚠️  More diverse training data (150K is small)
-  ⚠️  Longer training (2 epochs not enough)
-  ⚠️  Stronger regularization to prevent overfitting
-  ⚠️  Better evaluation on neutral benchmarks from day 1
 ```
 ---
-## ⚠️ Limitations
 - **Context window**: 512 tokens max
 - **English only**: Not suitable for other languages
@@ -255,20 +255,20 @@ What needs improvement:
 ---
-## 🎯 Intended Use
 | Use case | Suitable? |
 |---|---|
-| Research on adaptive computation | ✅ Yes |
-| Learning how LLMs are built from scratch | ✅ Yes |
-| Prototyping conversational AI | ✅ With caveats |
-| Production chatbots | ❌ No |
-| Medical / legal / financial advice | ❌ No |
-| Factual question answering | ❌ No |
 ---
-## 📄 Citation
 ```bibtex
 @misc{girinath2026mor,
@@ -285,7 +285,7 @@ What needs improvement:
 ---
-## 🙏 Acknowledgments
 - Anthropic — HH-RLHF dataset
 - Tsinghua University — UltraChat dataset
@@ -295,6 +295,6 @@ What needs improvement:
 ---
-**Model status** : ✅ Research / Educational use
 **Last updated** : March 2026
 **License**      : MIT

 ---
+## Novel Architecture: Mixture of Recursion (MoR)
 Traditional transformers apply the same depth to every input — simple and complex sentences both
 go through all N layers. This wastes compute on easy inputs and under-processes hard ones.
 ### What the Results Actually Show
 ```
+Architecture works   — training is stable, loss decreases
+Router works         — 0 NaN batches, routing signals valid
+Novel contribution   — perplexity-based self-supervised routing
+Generalization gap  — val PPL 15.37 vs test PPL 43.21
+More data needed    — 150K samples is small for this model size
 ```
 ---
+## Quick Start
 ```bash
 pip install transformers torch
 ---
+## Training Details
 ### Dataset — 150K Conversational Samples
 ---
+## Technical Innovation
 ### Self-Supervised Perplexity Routing
 ```
 What worked:
+  NaN-safe attention masking (-1e4)
+  Self-supervised routing signal
+  Stable FP16 training (0 NaN batches)
+  Perplexity drops meaningfully across epochs
 What needs improvement:
+  More diverse training data (150K is small)
+  Longer training (2 epochs not enough)
+  Stronger regularization to prevent overfitting
+  Better evaluation on neutral benchmarks from day 1
 ```
 ---
+## Limitations
 - **Context window**: 512 tokens max
 - **English only**: Not suitable for other languages
 ---
+## Intended Use
 | Use case | Suitable? |
 |---|---|
+| Research on adaptive computation | Yes |
+| Learning how LLMs are built from scratch | Yes |
+| Prototyping conversational AI | With caveats |
+| Production chatbots | No |
+| Medical / legal / financial advice | No |
+| Factual question answering | No |
 ---
+## Citation
 ```bibtex
 @misc{girinath2026mor,
 ---
+## Acknowledgments
 - Anthropic — HH-RLHF dataset
 - Tsinghua University — UltraChat dataset
 ---
+**Model status** : Research / Educational use
 **Last updated** : March 2026
 **License**      : MIT