User01110/CMA-5M-Point

SupraCMA 5M

Evaluated training checkpoint from a 5.09M-parameter Supra-style Channel-Mixing Attention generalist language model using the third-party AxiomicLabs GPT-S 4,096-token tokenizer. It has no place embeddings, role embeddings, or inference-time equation detection. It was recorded at step 20,000 with WikiText normalized BPB 1.4777.

Loading

This is a custom Transformers architecture. trust_remote_code=True is required because stock Hugging Face model classes do not implement CMA or this model's exact rotary convention.

pip install "torch>=2.5" "transformers>=4.50" safetensors

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

repo = "User01110/CMA-5M-Point"
device = "cuda" if torch.cuda.is_available() else "cpu"
dtype = torch.bfloat16 if device == "cuda" else torch.float32
tokenizer = AutoTokenizer.from_pretrained(repo, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
    repo,
    trust_remote_code=True,
    torch_dtype=dtype,
).to(device).eval()

inputs = tokenizer("The process of photosynthesis", return_tensors="pt").to(device)
output = model.generate(**inputs, max_new_tokens=64)
print(tokenizer.decode(output[0], skip_special_tokens=True))

Checkpoint tensors are stored in bfloat16. Pass torch_dtype=torch.float32 when an FP32 runtime is required; every stored BF16 value widens exactly to FP32, though the pre-export FP32 master-weight mantissa cannot be reconstructed. The model intentionally sets use_cache=False: generation is standard Transformers generation, but the visible context is recomputed for every new token because this experimental architecture does not implement a KV cache.

Architecture

• Parameters: 5,094,112, with tied input/output embeddings
• Weights: native bfloat16 safetensors (model.safetensors); no .bin weights
• Runtime: PyTorch 2.5+ native SDPA plus compiled dense CMA chunk routing
• Tokenizer: AxiomicLabs/GPT-S-5M at revision 275b9c3ca78736bf6aeb154c7e2d5f5764fe9035
• Vocabulary: 4,096 third-party tokens
• Parameter allocation: 884,736 tied embedding parameters and 4,209,376 non-embedding parameters
• Supported/exported context: 1,024 tokens
• Training block length: 1,024 tokens
• Standalone prompt tokenization automatically prepends the native BOS token
• Width/layers: 216 / 10
• Token-attention heads: 6 query, 2 KV
• CMA: 9 chunks of 24 channels, 3 routing heads, and 2x value expansion
• Queries combine local chunks, learned chunk identities, and a token-global projection; keys summarize the corresponding expanded value chunks
• Per-token softmax attention supplies dense all-to-all chunk interaction
• Routing is a signed residual correction: v + tanh(route_signal) * (attention(v) - v)
• The attention prior assigns 90% probability to each chunk's identity route before content logits
• Expanded routed values are modulated by a SiLU gate and projected back to the residual width; the output projection starts at zero for stable optimization
• Contiguous-half RoPE without scaling
• No task-specific model features or inference-time benchmark handling

Intended use and limitations

This is a small base language model released for architecture research, representation analysis, and controlled comparisons. It is not instruction tuned and should not be treated as a factual authority or used for consequential decisions. Its 5.09M-parameter scale, English-weighted web mixture, 1,024-token context, and cache-free generation materially limit capability and throughput. The usual web-corpus biases and inaccuracies remain.

Tokenizer provenance

The tokenizer and its 4,096-token vocabulary were not created or owned by the CMA model author. They are reused from the public AxiomicLabs/GPT-S-5M repository at the exact revision listed above, whose repository metadata identifies Axiomic Labs as the publisher and Apache-2.0 as the license. No claim of tokenizer ownership beyond that public attribution is made here. The exported copy preserves its vocabulary and tokenization pipeline; CMA only configures its existing <bos> token to be prepended automatically and raises the exported model/tokenizer context limit to 1,024 tokens.

Optimization

• Training budget: 15,728,640,000 tokens over 30,000 updates
• Effective batch: 524,288 tokens per update
• Native microbatch: 256 sequences x 1,024 tokens, accumulated 2 times
• Runtime implementation: explicit BF16 residual/CMA activations and compiled dense nine-chunk routing, with full-model torch.compile enabled by default
• Runtime: one uninterrupted 30,000-update session with exact full-precision optimizer/RNG/data-stream recovery checkpoints
• Learning rate: 750-update linear warmup, flat through 25,500, then cosine decay toward zero; AdamW peaks at 2.5e-03 and original-scaling Muon at 3.0e-02
• The final configured update is positive; update 30,001 is exactly zero
• Official PyTorch Muon with original matrix-shape scaling for hidden matrices; AdamW for embeddings and remaining parameters

Training mixture

• FineWeb-Edu 100BT shuffled: 55%
• Cosmopedia v2: 25%
• FineWeb-HQ: 10%
• FineMath 4+: 10%

FineWeb-Edu supplies the primary educational web text, Cosmopedia supplies synthetic textbook-style coverage, FineWeb-HQ contributes model-filtered, knowledge-rich general web text, and FineMath-4+ supplies mathematical reasoning as ordinary causal-language-model text. The mixture remains fixed for the complete run. There are no benchmark labels or benchmark-specific preprocessing. Dataset revisions are pinned and the same fixed mixture is maintained for the complete single-session run.

Zero-shot evaluation at step 20,000

The four lm-eval tasks use normalized accuracy when supplied by lm-eval 0.4.12, with native bfloat16 weights and float32 likelihood softmax. ArithMark-3 uses its official primary acc_norm metric: mean continuation-token log likelihood, with context and continuation tokenized separately after one native BOS prefix. Autocast is not used for evaluation. The four lm-eval benchmark contexts use the tokenizer's native BOS behavior.

Open SLM Intelligence Index: 5.76 Open SLM Average: 34.96%

Benchmark	Accuracy
HellaSwag	27.88%
ARC-Easy	33.29%
ARC-Challenge	24.06%
PIQA	54.57%
ArithMark-3 (acc_norm)	28.70%

The Average equally weights HellaSwag, combined ARC (the mean of ARC-Easy and ARC-Challenge), PIQA, and ArithMark-3. The Intelligence Index first maps random chance to 0 and perfect accuracy to 100, then weights HellaSwag, combined ARC, and PIQA at 1.0 and ArithMark-3 at 0.75, matching Open SLM Leaderboard revision b7781f0efb9df61912f94cf0e196ed32becded83.

WikiText-103 validation at this step: loss 3.2680, perplexity 26.26, normalized BPB 1.4777 over 359,037 scored tokens and 1,145,591 normalized UTF-8 bytes, using one initial BOS, 1,024-token windows, and a 512-token stride.