⚠️ Re-download if you pulled this bundle before 2026-07-28
Earlier uploads of this bundle were missing
top_k: 20ingeneration_config.jsonandjang_config.json. Poolside's ownLaguna-XS-2.1/generation_config.jsonomitstop_keven though their model card specifiestemperature=1.0, top_k=20, top_p=1for every XS-2.1 evaluation (the S-2.1 config does ship it), and this bundle passed those values through verbatim.Impact: with
top_kunset, a runtime samples the full 100,352-token vocabulary unfiltered attemperature=1.0/top_p=1.0instead of the top 20 candidates — noticeably lower output quality, worst on the lower-bit profiles.Fix: re-download
generation_config.jsonandjang_config.json(a few KB — no need to re-pull the weights), or passtop_k=20explicitly in your requests. Both files now carrytop_k: 20and are checked against each other at build time.
JANGQ-AI/Laguna-XS-2.1-JANG_6M
JANG_6M JANG affine bundle of poolside/Laguna-XS-2.1 — 33B-parameter / ~3B-active MoE for agentic coding, quantized for Apple Silicon.
Attention (incl. the softplus output gate), shared expert, router, and norms stay high precision; the 256-routed-expert bulk carries 6-bit gate/up/down, protected by activation-aware (AWQ) scales chosen by measured quantization error against a no-AWQ baseline (0.3% lower reconstruction error at this bit width).
Bundle
| Field | Value |
|---|---|
| Source | poolside/Laguna-XS-2.1 @ c405648 |
| Architecture | laguna — 40 layers (10 global + 30 SWA w512), 33B-A3B, 256 experts top-8 + shared, 262K ctx |
| On-disk size | 27.7 GB (6 shards) |
| Routed experts | 6-bit gate/up/down affine, group 64, AWQ folded |
| Attention q/k/v/o + g_proj | 8-bit affine |
| Shared expert / dense FFN | 8-bit affine |
| Embeddings / lm_head | 6-bit / 8-bit affine |
| Router, e_score bias, norms | fp16 passthrough |
| Modality | text-only (verified from tensor index) |
Measured (M5 Max, 128 GB, greedy, wired)
| Metric | Value |
|---|---|
| Decode | 68.6 tok/s |
| Long-context cache parity | teacher-forced top-1 agreement 0.998 / 0.998 (pre/post the 512 sliding window, 2,913-token pass) |
Chat / reasoning
- GLM-style think tags;
enable_thinkingtoggles reasoning (vendor serving default is ON (stamped injang_config.chat); the template's own fallback is off, so passenable_thinking=Trueexplicitly if your runtime ignores bundle defaults) - The FULL chat template is inlined in
tokenizer_config.json(upstream ships only an{% include %}stub that most runtimes cannot resolve — inlining is what makes the reasoning toggle actually work) - Stop tokens
eos_token_id = [2, 24]— id 24 is end-of-turn and must be in the stop set - Template emits its own leading
〈|EOS|〉(bos 2): do not prepend another - Tool calls:
<tool_call>name<arg_key>k</arg_key><arg_value>v</arg_value></tool_call> - Vendor sampling: temperature 1.0, top_p 1.0, top_k 20
Loaders must honor the per-module {bits, group_size, mode} overrides in config.json[quantization].
Quantized and verified by Jinho Jang (eric@jangq.ai). License: OpenMDW-1.1 (inherited).