Qwen3.6-35B-A3B-oQ3e-FP16-MTP-MLX

Qwen3.6-35B-A3B-oQ3e-FP16-MTP-MLX

This is the FP16 precision sibling of treeish/Qwen3.6-35B-A3B-oQ3e-MTP-MLX. It is intended for Sprig on M1 and M2 Macs, where FP16 avoids relying on BF16 arithmetic. M3 and newer Macs should use the original BF16 package.

The oQ3e quantised weights, embedded Multi-Token Prediction head, tokenizer and corrected chat template are all retained. This sibling changes storage precision only for tensors that were BF16 in the source package. It does not reduce the model's disk or memory footprint.

Model

• Base model: Qwen/Qwen3.6-35B-A3B
• Architecture: 35B total parameters, approximately 3B active parameters
• Quantisation: oQ3e imatrix-enhanced mixed precision
• Default quantisation: 3-bit affine, group size 64
• Per-tensor overrides: 5, 6 and 8-bit, with group sizes 64 and 128
• Residual floating-point precision: FP16
• Format: MLX safetensors
• MTP: 1 embedded layer under language_model.mtp.*
• Context length: 262,144 tokens

The package contains 2,052 indexed tensors, including 42 embedded MTP tensors. It occupies essentially the same 17.2 GB as its BF16 sibling.

Conversion

The source is treeish/Qwen3.6-35B-A3B-oQ3e-MTP-MLX at commit 85dd79d1f9169cf5c7c7c63e4fb2658680e1ed8c.

The conversion used MLX 0.29.3 and follows the FP16 precision-sibling policy of youssofal/MTPLX at commit 21be78b3f51820eecef020e5e4855c0715eaf9a5, with a check of the output against the source:

• Cast every BF16 tensor to FP16.
• Preserve packed quantised tensors and every other non-BF16 tensor.
• Preserve safetensors metadata and copy non-weight files byte for byte before updating release documentation.
• Compare every output tensor with the expected converted or preserved source value.

The conversion changed 1,569 BF16 tensors to FP16 and preserved 483 non-BF16 tensors. 333 of 333 vision-tower tensors became FP16. Of the 42 embedded MTP tensors, 31 became FP16 and the 11 packed tensors were preserved. FP16_CONVERSION_MANIFEST.json records every tensor, dtype, shape and source/output shard digest.

The parent package's model provenance and Froggeric chat-template attribution remain applicable. See its model card and this package's RELEASE_MANIFEST.json for the pinned upstream revisions.

Runtime compatibility

This package is built for Treeish's pinned MLX Swift runtime. Another runtime must support the per-tensor quantisation overrides in config.json and the embedded Qwen MTP layout.

This package is Treeish's M1 and M2 variant. M3 and newer Macs use the BF16 sibling.

Treeish uses this model from 32 GB of unified memory and recommends 48 GB. Headroom depends on context length, cache settings and other running applications.

Treeish validation

Treeish's release benchmark has not yet been run on this package. RELEASE_MANIFEST.json records the release status and is updated when it has.

Limitations

Quantisation trades some model quality for memory use and local generation speed. The FP16 sibling changes residual floating-point precision, not quantisation or active parameter count.

The model package contains no custom executable code. File sizes, SHA-256 digests and source revisions are recorded in RELEASE_MANIFEST.json and FP16_CONVERSION_MANIFEST.json.

Licence

Qwen3.6-35B-A3B is licensed under Apache 2.0. The full licence text is included in LICENSE. The Froggeric template repository and MTPLX also declare Apache 2.0 and are attributed above.