GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
tablinum
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
tablinum
Model Tree
moonshotai/Moonlight-16B-A3B-Instruct
Input Modalities
Output Modalities
Supported Functionality
Moon-Beam is a private reference desk that runs on your own computer. Give it the Tablinum-2 knowledge library and it answers factual and how-to questions from 26 tables of reference text (Wikipedia, law and regulation, software documentation, home repair and DIY Q&A, philosophy, psychology, open textbooks, research papers), offline. When the passages don't hold the answer, it says it isn't sure instead of making one up.
Good uses:
llama-server, mlx_lm.server) plus the retrieval step.Not for:
One quirk to know: asked "can you search the web?" or "can you run this code?", Moon-Beam 16B sometimes says yes. It can't. It only has its training and the passages you give it. We tested default system prompts that tell it so, and every one cost accuracy on easy facts (20 of 24 right with the original prompt, 13 to 17 with the new ones), so the chat template is unchanged. Moon-Beam 8B is trained to answer questions about itself correctly.
The knowledge library is a separate download today (Tablinum-2). Without it Moon-Beam is a careful general chat model; with it, it is the reference desk above. A one-command setup that installs both together is coming with the next release.
| Held-out final (IQ3, paired) | untrained Moonlight | Moon-Beam | right / wrong with engram |
|---|---|---|---|
| EngramBench 502, all 26 tables | −58.6 → +65.1 | −46.8 → +73.3 | 406 / 38 (untrained 406 / 79) |
| SimpleQA-verified 1000 | −88.6 → +17.7 | −36.0 → +34.9 | 504 / 155 (untrained 548 / 371) |
| AA-Omniscience test 400 | −75.5 → −42.5 | −57.2 → −16.5 | 78 / 144 (untrained 96 / 266) |
The index is 100 × (correct − incorrect) / n, so a wrong answer costs as much as a right one earns. The engram supplies the knowledge. The training adds calibration: with the passages in hand Moon-Beam finds the answer about as often as its base, but it stops guessing when the passages don't hold it. All arms are graded by Claude Haiku against the gold answers, paired on the same questions; questions the grader declined to grade are dropped for both models. None of these questions, or any passage retrieved for them, was used in training.
EngramBench is ours: 24 questions per subject, written from passages held out of training, each with a short gold answer that appears in its passage. One retrieval over all 26 tables finds the right subject's passage in the top 3 for 95% of questions, so no per-domain router is needed. Searching everything costs about one index point against sending each question to its own table.
The merged BF16 weights in the original Moonlight layout: 27 safetensors shards (31.9 GB), the Moonlight tokenizer (tiktoken.model)
and its remote code. Quantized builds for llama.cpp are in GGUF files:
| GGUF | size | KLD vs these weights (chat / wiki) | same top-1 | EngramBench own-table | SimpleQA + engram |
|---|---|---|---|---|---|
Moon-Beam-16B-A3B-IQ2_XXS.gguf | 6.08 GB | 0.114 / 0.235 | 89.1% | +73.7 | +31.6 |
Moon-Beam-16B-A3B-IQ3_XXS.gguf (recommended) | 8.06 GB | 0.027 / 0.061 | 95.0% | +74.1 | +34.9 |
Moon-Beam-16B-A3B-IQ4_XS.gguf | 9.67 GB | 0.006 / 0.015 | 97.5% | +71.7 | +37.7 |
Moon-Beam is a QLoRA fine-tune: rank 16 on the MLA attention, the shared experts and the dense layer, with the 64 routed experts frozen. It trained one epoch on one 12 GB GPU, on 2,198 examples written by Claude Haiku from passages of the 21 non-health Tablinum-2 tables:
Health tables never left the machine. It took three rounds: the pilot learned to decline everything, and round two answered everything briefly and confidently. Round three, the recipe above, keeps both the help and the honesty.
| general skill (IQ3, same prompt for both) | untrained Moonlight | Moon-Beam |
|---|---|---|
| GSM8K (400, short chain of thought) | 83.5% | 71.8% |
| MMLU-Pro (700, 14 × 50, zero-shot letter) | 36.9% | 31.7% (35.8% on the 619 it answered with a letter) |
The training data is short factual and how-to answers, and it costs some general skill. Moon-Beam's math working is shorter and it skips steps more often, and it sometimes answers a multiple-choice question in a sentence instead of a letter. On the questions it does answer with a letter, MMLU-Pro knowledge is unchanged. If you need multi-step math, use the base model or route math to a tool.
python
from transformers import AutoModelForCausalLM, AutoTokenizertok = AutoTokenizer.from_pretrained("tablinum/Moon-Beam-16B-A3B", trust_remote_code=True)model = AutoModelForCausalLM.from_pretrained("tablinum/Moon-Beam-16B-A3B", torch_dtype="auto", device_map="auto", trust_remote_code=True)msgs = [{"role": "user", "content": "..."}]ids = tok.apply_chat_template(msgs, add_generation_prompt=True, return_tensors="pt").to(model.device)print(tok.decode(model.generate(ids, max_new_tokens=512)[0][ids.shape[1]:], skip_special_tokens=True))The tokenizer needs trust_remote_code (Moonlight's tiktoken tokenizer); recent transformers versions also load the model natively as
DeepseekV3ForCausalLM. For llama.cpp use the GGUF repo.
Use the v1 injection prompt with retrieved passages:
Reference passages (retrieved automatically; they may be incomplete or irrelevant): … Use a passage only if it states the answer to this exact question. A passage about a related but different item, version, year, rule or case does not answer it: then rely on your own knowledge, or say you don't know.
MIT. Moonlight-16B-A3B-Instruct is by Moonshot AI (MIT). Moon-Beam is a fine-tuned derivative by Lightspeed / Tablinum. Keep both copyright notices with any redistribution. Training examples were written with Claude Haiku, and the character art is Tablinum's own. No endorsement by Moonshot AI or Anthropic is implied.
bibtex
@misc{moonbeam16b2026, title = {Moon-Beam 16B-A3B}, author = {Lightspeed / tablinum}, year = {2026}, note = {Moonlight-16B-A3B fine-tuned to use the Tablinum-2 engram}, url = {https://huggingface.co/tablinum/Moon-Beam-16B-A3B}}from transformers import AutoModelForCausalLM, AutoTokenizertok = AutoTokenizer.from_pretrained("tablinum/Moon-Beam-16B-A3B", trust_remote_code=True)model = AutoModelForCausalLM.from_pretrained("tablinum/Moon-Beam-16B-A3B", torch_dtype="auto", device_map="auto", trust_remote_code=True)msgs = [{"role": "user", "content": "..."}]ids = tok.apply_chat_template(msgs, add_generation_prompt=True, return_tensors="pt").to(model.device)print(tok.decode(model.generate(ids, max_new_tokens=512)[0][ids.shape[1]:], skip_special_tokens=True))@misc{moonbeam16b2026, title = {Moon-Beam 16B-A3B}, author = {Lightspeed / tablinum}, year = {2026}, note = {Moonlight-16B-A3B fine-tuned to use the Tablinum-2 engram}, url = {https://huggingface.co/tablinum/Moon-Beam-16B-A3B}}