Text Generation
Transformers
Safetensors
English
gabgpt
gift-of-gab
small-language-model
trained-from-scratch
byte-level
looped-transformer
browser
conversational
custom_code
Instructions to use gszauer/Gab10M with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use gszauer/Gab10M with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="gszauer/Gab10M", trust_remote_code=True) 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("gszauer/Gab10M", trust_remote_code=True, device_map="auto") - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- vLLM
How to use gszauer/Gab10M with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "gszauer/Gab10M" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "gszauer/Gab10M", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker
docker model run hf.co/gszauer/Gab10M
- SGLang
How to use gszauer/Gab10M 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 "gszauer/Gab10M" \ --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": "gszauer/Gab10M", "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 "gszauer/Gab10M" \ --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": "gszauer/Gab10M", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }' - Docker Model Runner
How to use gszauer/Gab10M with Docker Model Runner:
docker model run hf.co/gszauer/Gab10M
Download model.json from gszauer/Gab10M: direct link, hf CLI and curl.
- Browser
- Download file 2.3 kB
-
https://huggingface.co/gszauer/Gab10M/resolve/main/model.json
- Command line
-
hf download hf://gszauer/Gab10M/model.json
-
curl -L -o model.json https://huggingface.co/gszauer/Gab10M/resolve/main/model.json
2.3 kB
| { | |
| "name": "Gab10M", | |
| "architecture": "gab10m", | |
| "vocabSize": 256, | |
| "featureDim": 256, | |
| "numHeads": 4, | |
| "ropeBase": 10000, | |
| "numPreludeBlocks": 2, | |
| "numCoreBlocks": 9, | |
| "numCodaBlocks": 2, | |
| "loopCount": 2, | |
| "mlpHiddenDim": 680, | |
| "convKernel": 4, | |
| "contextLength": 2048, | |
| "derived": { | |
| "headDim": 64, | |
| "parameterCount": 10302720, | |
| "uniqueBlocks": 13, | |
| "effectiveDepth": 22 | |
| }, | |
| "looping": { | |
| "schedule": "prelude once, core loopCount times with shared weights, coda once", | |
| "injection": "additive: the prelude output is added to the residual stream at the start of every core pass after the first", | |
| "kvCache": "one slot per layer pass (effectiveDepth slots), so each core block has loopCount slots; each slot also holds that pass's two canon histories" | |
| }, | |
| "chat": { | |
| "user": "<|u|>", | |
| "assistant": "<|a|>", | |
| "end": "<|e|>", | |
| "endOfText": "<|endoftext|>", | |
| "encoding": "markers are plain UTF-8 text spelled out byte by byte; no reserved ids", | |
| "template": "<|u|>{user}<|e|><|a|>{assistant}<|e|>, repeated for each turn pair", | |
| "structure": "one or more user/assistant turn pairs, starting with the user; no system prompt", | |
| "generationPrompt": "the conversation so far followed by <|a|>; the reply ends at <|e|>" | |
| }, | |
| "weights": { | |
| "format": "self-describing GAB_ file: 12-byte preamble (magic GAB_, version, header length), JSON header, then little-endian float32 weights in this order", | |
| "matrixLayout": "row-major [in, out]; activations are multiplied as x @ W", | |
| "order": [ | |
| "tokenEmbeddings [vocabSize, featureDim] (tied: also the output projection, transposed)", | |
| "for each unique block, prelude then core then coda: attentionNorm.gamma [1, featureDim], canonA.weight [convKernel, featureDim] (row j weighs the input j positions back)", | |
| " for each head: learnedQ [featureDim, headDim], learnedK [featureDim, headDim], learnedV [featureDim, headDim], learnedO [headDim, featureDim], queryNorm.gamma [1, headDim], keyNorm.gamma [1, headDim]", | |
| " mlpNorm.gamma [1, featureDim], canonC.weight [convKernel, featureDim], mlp.learnedGate [featureDim, mlpHiddenDim], mlp.learnedUp [featureDim, mlpHiddenDim], mlp.learnedDown [mlpHiddenDim, featureDim]", | |
| "finalNorm.gamma [1, featureDim]" | |
| ] | |
| } | |
| } | |