Instructions to use tablinum/Lite-Beam-27B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use tablinum/Lite-Beam-27B with llama.cpp:

Install (macOS, Linux)

curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf tablinum/Lite-Beam-27B:IQ1_M
# Run inference directly in the terminal:
llama cli -hf tablinum/Lite-Beam-27B:IQ1_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf tablinum/Lite-Beam-27B:IQ1_M
# Run inference directly in the terminal:
llama cli -hf tablinum/Lite-Beam-27B:IQ1_M

Build from source code

git clone https://github.com/ggml-org/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-app
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama serve -hf tablinum/Lite-Beam-27B:IQ1_M
# Run inference directly in the terminal:
./build/bin/llama cli -hf tablinum/Lite-Beam-27B:IQ1_M

Use Docker

docker model run hf.co/tablinum/Lite-Beam-27B:IQ1_M

Llama new
LM Studio
Jan

vLLM

How to use tablinum/Lite-Beam-27B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "tablinum/Lite-Beam-27B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "tablinum/Lite-Beam-27B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/tablinum/Lite-Beam-27B:IQ1_M

Ollama
How to use tablinum/Lite-Beam-27B with Ollama:
```
ollama run hf.co/tablinum/Lite-Beam-27B:IQ1_M
```
Unsloth Desktop

How to use tablinum/Lite-Beam-27B with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf tablinum/Lite-Beam-27B:IQ1_M

Configure the model in Pi

# Install Pi:
npm install -g @earendil-works/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "tablinum/Lite-Beam-27B:IQ1_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use tablinum/Lite-Beam-27B with Docker Model Runner:
```
docker model run hf.co/tablinum/Lite-Beam-27B:IQ1_M
```

Lemonade

How to use tablinum/Lite-Beam-27B with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull tablinum/Lite-Beam-27B:IQ1_M

Run and chat with the model

lemonade run user.Lite-Beam-27B-IQ1_M

List all available models

lemonade list

Hermes Agent

How to use tablinum/Lite-Beam-27B with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf tablinum/Lite-Beam-27B:IQ1_M

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default tablinum/Lite-Beam-27B:IQ1_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use tablinum/Lite-Beam-27B with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf tablinum/Lite-Beam-27B:IQ1_M

Configure OpenClaw

# Install OpenClaw:
npm install -g openclaw@latest
# Register the local server and set it as the default model:
openclaw onboard --non-interactive --mode local \
  --auth-choice custom-api-key \
  --custom-base-url http://127.0.0.1:8080/v1 \
  --custom-model-id "tablinum/Lite-Beam-27B:IQ1_M" \
  --custom-provider-id llama-cpp \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

Lite-Beam 27B

Qwen3.8-27B in 9.26 GB (2-bit) or 6.73 GB (1-bit) GGUFs that run on small consumer GPUs. A per-model gated engram lifts knowledge far past where the quant starts, and the 1-bit build gains the most. With Tablinum-2, the 1-bit build scores +43.0 on SimpleQA (only 7 wrong of 200) and +13.5 on AA-Omniscience dev, up from −14.5. Both are the best results of the builds we tested. Apache-2.0.

This is a quantized derivative of Qwen/Qwen3.8-27B, not a new base model. The weights are built with Tablinum's calibration set, importance matrix and quantizer (the Lite-Beam method), on per-tensor type layouts credited below. Two files: IQ1_M (6.73 GB, 1-bit), the recommended build when paired with the engram, and IQ2_S (9.26 GB, 2-bit), the stronger build without retrieval. Use the v1 injection prompt (below) with either.

Quick links: Files · Run locally · Tablinum-2 engram · Lite-Beam 1.0 (501B)

The result

Lite-Beam 27B engram scorecard: closed-book vs Tablinum-2 engram on AA-Omniscience, SimpleQA, gated AA dev, GPQA and AIME

Knowledge (held-out)	build	closed-book	+ engram	Δ	correct / incorrect with engram
SimpleQA-verified (dev 200)	IQ1_M 6.73 GB	−14.0	+43.0	+57.0	93 / 7
	IQ2_S 9.26 GB	−25.5	+38.5	+64.0	99 / 22
AA-Omniscience (test 400), plain prompt	IQ1_M	−7.2	−3.5	+3.7	72 / 86
	IQ2_S	−19.8	−1.0	+18.8 (93 better / 36 worse, p = 5·10⁻⁷)	115 / 119

The index is 100 × (correct − incorrect) / n, so a wrong answer costs as much as a right one earns. The engram mostly helps by replacing confident wrong answers with grounded ones. The 1-bit build starts better calibrated (it declines instead of guessing: 71 wrong vs 149 on AA closed-book) and, with the v1 prompt, keeps that discipline when passages are injected.

These test-split arms use the Software + Law tables with the plain injection prompt on AA, and the Wikipedia table on SimpleQA. They are self-judged at low reasoning effort, so the absolute values are not comparable with published AA numbers; the paired deltas are the result. The release route is the per-domain gate with the v1 prompt (below).

Per-domain gate (AA dev 200)

Each table is kept for a model only if it helps that model in a paired dev test. v1 is the injection prompt that says: use a passage only if it states the answer to this exact question; otherwise rely on your own knowledge or say you don't know.

IQ1_M (6.73 GB):

domain	closed-book	engram, plain	engram, `v1`	better / worse (`v1`)	gate
Software	−23.5	+47.1	+44.1	20 / 1 (p < .001)	on
Law	0.0	+15.2	+21.2	10 / 3	on, `v1`
STEM	−15.2	−9.1	+15.2	13 / 4 (p = .049)	on, `v1` only
Finance	−12.1	+6.1	+3.0	8 / 3	on
Health	−36.4	−3.0	−6.1	11 / 4	on
Humanities	0.0	−2.9	+2.9	3 / 3	flat
all 200	−14.5	+9.0	+13.5		best per domain ≈ +15

IQ2_S (9.26 GB): closed-book −32.5 → +1.5 with the plain prompt; Health +6.1 with v1; gated ≈ +3.

build	closed-book	gated engram	gain
IQ1_M 6.73 GB	−14.5	≈ +15	+30
IQ2_S 9.26 GB	−32.5	≈ +3	+35

The 1-bit build starts highest because it guesses less, gains about as much from the engram, and with v1 is the only build where STEM tables help. The simple route, v2 tables with v1 on every domain, scores +13.5 with no per-domain tuning. STEM and Humanities remain coverage-limited (the answer is in the top 3 passages for few questions), and better sources for both are being built for Tablinum-2.

Reasoning

benchmark	build	score
GPQA Diamond (198), low effort, 32K budget	IQ2_S 9.26 GB	139/198 (70.2%)

10 answers still reached the 32K budget and count as wrong; on the 188 that finished, the score is 73.9%. The 1-bit build's reasoning runs were not made for this release.

What changed

	IQ2_S (2-bit)	IQ1_M (1-bit)
size	9,259,511,744 bytes (8.62 GiB)	6,729,166,784 bytes (6.27 GiB)
bpw	2.71	1.97
KLD vs BF16, chat (assistant tokens)	0.121	0.418
KLD vs BF16, wikitext	0.174	0.475
same top-1 token, chat	—	85.8%
imatrix / calibration	Tablinum	Tablinum
per-tensor layout	ISTA-DASLab GSQ-RCO IQ2_S	Unsloth UD-IQ1_M

KLD is measured only on assistant-turn tokens: an instruct model's system and user tokens are untrained and swamp the mean otherwise.

IQ1_M KLD was measured on the CPU; IQ2_S on the GPUs. The 1-bit build keeps 86% top-1 agreement with BF16, but its closed-book knowledge drops more, which the engram is there to recover.

IQ2_S map: attention IQ3_S / IQ3_XXS, FFN mostly IQ2_XXS–IQ3_XXS with a few IQ1 bands, output IQ4_XS, token embeddings IQ2_S, DeltaNet gates F32/BF16. The exact per-tensor maps are in the two manifests.

Files

file	bytes	size	use
`Lite-Beam-27B-IQ2_S.gguf`	9,259,511,744	8.62 GiB	llama.cpp
`Lite-Beam-27B-IQ1_M.gguf`	6,729,166,784	6.27 GiB	llama.cpp, smallest machines
`LICENSE`	—	—	Apache-2.0
`NOTICE`	—	—	attribution
`quantization_manifest.json`	—	—	IQ2_S: exact map, KLD, hash
`quantization_manifest_IQ1_M.json`	—	—	IQ1_M: exact map, KLD, hash
`benchmarks/`	—	—	graded arms and scripts

SHA-256:

Lite-Beam-27B-IQ2_S.gguf f08759290b3b117d3f8a7f33914148ec13b0186c6c791a5b86eaad53c362ac63
Lite-Beam-27B-IQ1_M.gguf a2c792dc720efd4077cd1cf4c903208a924947c653b2f3188472eb979ea0ca42

Run locally

Reference rig: RTX A2000 12 GB + RTX 2060 6 GB, split 7:2.

llama-server -m Lite-Beam-27B-IQ2_S.gguf --alias lite-beam-27b \
  -ngl 99 -ts 7,2 -fa on -np 4 -c 57344 -ctk q8_0 -ctv q8_0 \
  --jinja --host 127.0.0.1 --port 18371

speed (A2000 + 2060)	t/s
prompt 512	361
prompt 4096	430
generation	14.6
generation with EAGLE-4 draft (1 slot)	23.6 (19.8–27.4)

Faster single-stream: the EAGLE-4 draft for Qwen3.8-27B works with this quant, at +63% generation speed on our rig and close to the reference acceptance rate. It needs a llama.cpp fork with EAGLE-4 support; see quimmedes/Qwen3.8-27B-EAGLE4-GGUF. The draft is under the Eagle 4 source-available license, not this repo's, and is not bundled.

Tablinum-2 engram

The gated retrieval tables, routes and code are in tablinum/tablinum-2. Tables whose sources do not allow redistribution are shipped as exact build recipes, not text. The engram runs locally; nothing is sent to a remote endpoint.

What you need

About 10 GB of VRAM across one or two GPUs for the 9.26 GB file with a 57K context, or about 8 GB for the 1-bit file. CPU offload works but is slow.
A recent llama.cpp with Qwen3.5/3.8 (qwen35) support and --jinja.
Optional: the Tablinum-2 tables (sizes vary per domain) and its retrieval code.
For maximum accuracy use the BF16 / FP8 checkpoint. This is the 2-bit local build.

License and attribution

Qwen3.8-27B is by the Qwen team (Alibaba Cloud), Apache-2.0. This GGUF is a modified, quantized derivative produced by Lightspeed / Tablinum. The per-tensor type layout follows ISTA-DASLab/Qwen3.8-27B-GSQ-RCO-GGUF (Apache-2.0); the tensor data is our own quantization. The IQ1_M build follows the per-tensor type layout of unsloth/Qwen3.8-27B-GGUF UD-IQ1_M (Apache-2.0); again the tensor data is our own. Include the Apache-2.0 text and the NOTICE with any redistribution. No endorsement by Qwen, ISTA-DASLab or Unsloth is implied.

Citation

@misc{litebeam27b2026,
  title  = {Lite-Beam 27B},
  author = {Lightspeed / tablinum},
  year   = {2026},
  note   = {Mixed 2-bit GGUF of Qwen/Qwen3.8-27B with a per-model gated engram},
  url    = {https://huggingface.co/tablinum/Lite-Beam-27B}
}