cstr/maia3-onnx-int32

Maia3 ONNX Models (int32 ELO inputs)

Maia3 neural network models with int32 ELO inputs for Safari/WebKit compatibility.

Original models use int64 for self_elo and oppo_elo inputs, which Safari's WebAssembly doesn't support (no BigInt64Array). These models add Cast nodes (int32 → int64) at the input, allowing int32 tensors while keeping internal computation unchanged.

Models

File	Size	Parameters
maia3_5m_int32.onnx	~22MB	5M
maia3_23m_int32.onnx	~92MB	23M
maia3_79m_int32.onnx	~313MB	79M

Tensor Specs

• Input: tokens — [batch, 64, 96] float32
• Input: self_elo — [1] int32 (player ELO, 0-5000)
• Input: oppo_elo — [1] int32 (opponent ELO, 0-5000)
• Output: logits_move — [batch, 4352] float32 (policy logits)
• Output: logits_value — [batch, 3] float32 (loss, draw, win)

Source

Original models from cemoss17/maia3-onnx. Modified with onnx Cast nodes for int32 input compatibility.

Used by CrispChess (MIT licensed chess app).