Instructions to use MatterWaveAI/grt-30m-chat with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use MatterWaveAI/grt-30m-chat with Transformers:

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

pipe = pipeline("text-generation", model="MatterWaveAI/grt-30m-chat")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("MatterWaveAI/grt-30m-chat", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use MatterWaveAI/grt-30m-chat with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "MatterWaveAI/grt-30m-chat"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "MatterWaveAI/grt-30m-chat",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/MatterWaveAI/grt-30m-chat

SGLang

How to use MatterWaveAI/grt-30m-chat 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 "MatterWaveAI/grt-30m-chat" \
    --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": "MatterWaveAI/grt-30m-chat",
		"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 "MatterWaveAI/grt-30m-chat" \
        --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": "MatterWaveAI/grt-30m-chat",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use MatterWaveAI/grt-30m-chat with Docker Model Runner:
```
docker model run hf.co/MatterWaveAI/grt-30m-chat
```

grt-30m-chat / eval_results.json

davidhouseholter

Upload MatterWaveAI/grt-30m-chat release checkpoint and model card

ffdafea verified 1 day ago

Raw History Blame Contribute Delete

3.05 kB

{
  "model_id": "MatterWaveAI/grt-30m-chat",
  "architecture": "GRTForCausalLM",
  "parameters": 29305498,
  "nn_linear_layers": 0,
  "chat_template": "ChatML (<|im_start|> / <|im_end|>)",
  "before_sft": {
    "val_loss": 5.2083,
    "val_ppl": 182.7854
  },
  "after_sft": {
    "val_loss": 0.3129,
    "val_ppl": 1.3674,
    "samples": [
      {
        "user": "Can you explain this briefly? What is GRT-30M?",
        "target": "GRT-30M is a 29.3M-parameter Givens Rotational Transformer developed by MatterWave AI that replaces every dense nn.Linear matrix with cascaded 2D Givens rotations and butterfly permutations.",
        "generated": "GRT-30M is a 29.3M-parameter Givens Rotational Transformer developed by MatterWave AI that replaces every dense nn.Linear matrix with cascaded 2D Givens rotations and butterfly permutations."
      },
      {
        "user": "Can you explain this briefly? How does a Givens Rotational Transformer replace dense linear layers?",
        "target": "Instead of O(D^2) dense matrix multiplications, GRT represents hidden states as planar coordinate pairs and transforms them through butterfly-permuted 2D Givens rotations modulated by Clifford algebra features.",
        "generated": "Instead of O(D^2) dense matrix multiplications, GRT represents hidden states as planar coordinate pairs and transforms them through butterfly-permuted 2D Givens rotations modulated by Clifford algebra features."
      },
      {
        "user": "Can you explain this briefly? What is PearlNorm in the GRT architecture?",
        "target": "PearlNorm is a pair-aware RMS normalization layer that divides each planar (x, y) pair by the root-mean-square of pair radii, rescaling magnitudes while preserving every pair's phase angle.",
        "generated": "PearlNorm is a pair-aware RMS normalization layer that divides each planar (x, y) pair by the root-mean-square of pair radii, rescaling magnitudes while preserving every pair's phase angle."
      },
      {
        "user": "Can you explain this briefly? What is RadialGELU?",
        "target": "RadialGELU applies a GELU activation to the Euclidean radius r of each 2D coordinate pair as gelu(r) / r, introducing nonlinearity on pair magnitudes without distorting phase directions.",
        "generated": "RadialGELU applies a GELU activation to the Euclidean radius r of each 2D coordinate pair as gelu(r) / r, introducing nonlinearity on pair magnitudes without distorting phase directions."
      },
      {
        "user": "Can you explain this briefly? How are token embeddings parameterized in GRT?",
        "target": "Each token in GRTEmbedding is stored as hidden_size // 2 phase angles plus a scalar log-magnitude, emitting exp(log_magnitude) * (cos(theta), sin(theta)) on the hypersphere.",
        "generated": "Each token in GRTEmbedding is stored as hidden_size // 2 phase angles plus a scalar log-magnitude, emitting exp(log_magnitude) * (cos(theta), sin(theta)) on the hypersphere."
      }
    ]
  }
}

	{
	"model_id": "MatterWaveAI/grt-30m-chat",
	"architecture": "GRTForCausalLM",
	"parameters": 29305498,
	"nn_linear_layers": 0,
	"chat_template": "ChatML (<\|im_start\|> / <\|im_end\|>)",
	"before_sft": {
	"val_loss": 5.2083,
	"val_ppl": 182.7854
	},
	"after_sft": {
	"val_loss": 0.3129,
	"val_ppl": 1.3674,
	"samples": [
	{
	"user": "Can you explain this briefly? What is GRT-30M?",
	"target": "GRT-30M is a 29.3M-parameter Givens Rotational Transformer developed by MatterWave AI that replaces every dense nn.Linear matrix with cascaded 2D Givens rotations and butterfly permutations.",
	"generated": "GRT-30M is a 29.3M-parameter Givens Rotational Transformer developed by MatterWave AI that replaces every dense nn.Linear matrix with cascaded 2D Givens rotations and butterfly permutations."
	},
	{
	"user": "Can you explain this briefly? How does a Givens Rotational Transformer replace dense linear layers?",
	"target": "Instead of O(D^2) dense matrix multiplications, GRT represents hidden states as planar coordinate pairs and transforms them through butterfly-permuted 2D Givens rotations modulated by Clifford algebra features.",
	"generated": "Instead of O(D^2) dense matrix multiplications, GRT represents hidden states as planar coordinate pairs and transforms them through butterfly-permuted 2D Givens rotations modulated by Clifford algebra features."
	},
	{
	"user": "Can you explain this briefly? What is PearlNorm in the GRT architecture?",
	"target": "PearlNorm is a pair-aware RMS normalization layer that divides each planar (x, y) pair by the root-mean-square of pair radii, rescaling magnitudes while preserving every pair's phase angle.",
	"generated": "PearlNorm is a pair-aware RMS normalization layer that divides each planar (x, y) pair by the root-mean-square of pair radii, rescaling magnitudes while preserving every pair's phase angle."
	},
	{
	"user": "Can you explain this briefly? What is RadialGELU?",
	"target": "RadialGELU applies a GELU activation to the Euclidean radius r of each 2D coordinate pair as gelu(r) / r, introducing nonlinearity on pair magnitudes without distorting phase directions.",
	"generated": "RadialGELU applies a GELU activation to the Euclidean radius r of each 2D coordinate pair as gelu(r) / r, introducing nonlinearity on pair magnitudes without distorting phase directions."
	},
	{
	"user": "Can you explain this briefly? How are token embeddings parameterized in GRT?",
	"target": "Each token in GRTEmbedding is stored as hidden_size // 2 phase angles plus a scalar log-magnitude, emitting exp(log_magnitude) * (cos(theta), sin(theta)) on the hypersphere.",
	"generated": "Each token in GRTEmbedding is stored as hidden_size // 2 phase angles plus a scalar log-magnitude, emitting exp(log_magnitude) * (cos(theta), sin(theta)) on the hypersphere."
	}
	]
	}
	}