Instructions to use MemTensor/MemT-1.0-Flash with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use MemTensor/MemT-1.0-Flash with Transformers:

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

pipe = pipeline("text-generation", model="MemTensor/MemT-1.0-Flash", trust_remote_code=True)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("MemTensor/MemT-1.0-Flash", trust_remote_code=True, device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use MemTensor/MemT-1.0-Flash with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "MemTensor/MemT-1.0-Flash"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "MemTensor/MemT-1.0-Flash",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/MemTensor/MemT-1.0-Flash

SGLang

How to use MemTensor/MemT-1.0-Flash 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 "MemTensor/MemT-1.0-Flash" \
    --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": "MemTensor/MemT-1.0-Flash",
		"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 "MemTensor/MemT-1.0-Flash" \
        --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": "MemTensor/MemT-1.0-Flash",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use MemTensor/MemT-1.0-Flash with Docker Model Runner:
```
docker model run hf.co/MemTensor/MemT-1.0-Flash
```

Disentangling Gain Sources in KDA and Two-Gate Gated Delta Variants

by SDFHSIDFdFEF - opened 6 days ago

Discussion

SDFHSIDFdFEF

6 days ago

What’s the main difference from KDA here? They’re both on the gated delta rule, and this one looks like it adds two LSTM-style gates. But could the gain mostly come from the retention post-training rather than the architecture? Have you tried KDA as the update rule in the same framework, so the two can be separated?

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment