Model card: the exact command that made the file

README.md

@@ -30,7 +30,7 @@ with the model's MTP head for speculative decoding and its vision tower.
 | File | `qwen3.8-next-flash-fp8-iq4r-moe.rad` — 113.6 GiB |
 | Routed experts | 4-bit codes into a 16-level non-uniform codebook after a 128-point Walsh–Hadamard rotation, one E4M3 scale per 64 weights; codes chosen by GPTQ (10M-token calibration, per-expert down-projection Hessians) and refined by three sweeps of coordinate descent |
 | Protected experts | ten experts that carry most of their layer's down-projection energy, kept bf16 |
-| Trunk | attention and shared-expert linears and the lm_head int8 (W8A8), a scale per 128 columns searched for least error; hyper-connection mixing matrices E4M3 |
+| Trunk | attention, delta-net and shared-expert linears and the lm_head int8 (W8A8), a scale per 128 columns searched for least error; hyper-connection mixing matrices E4M3 |
 | Speculator | the model's MTP head (depth 3 by default) |
 | Vision | the vision tower, bf16: images and video in chat requests |
 | Context | 262,144 tokens trained; served at 200K |
@@ -65,6 +65,63 @@ tokens per step, and ~6,400 tokens/s prefill on a 26K-token prompt.
 The server speaks the OpenAI API (`/v1/chat/completions`, `/v1/completions`), with tool calls,
 structured output, and `image_url` / video parts in chat messages.
 
+## How this file was made
+
+The exact command, run on 2026-10-03 from a radiance build directory of the commit series merged
+to radiance's main branch as 2531f14 (`radiance_home` is that build's plugin directory):
+
+```sh
+CALIB=~/models/calib/w4nl-calib OMP_NUM_THREADS=12 nice -n 10 ./bin/rad-convert \
+    ~/models/Qwen/Qwen3.8-Flash-Next --recipe data/recipes/qwen4exp-w4nl64-i8-hc8m.recipe \
+    --home radiance_home --tokenizer ~/models/Qwen/Qwen3.8-Flash-Next/tokenizer.json \
+    --reuse ~/models/rad/w4nl-i8lm.rad -o ~/models/rad/w4nl64-i8lm.rad -v
+```
+
+- `--reuse` names an earlier container converted from the same checkpoint by an intermediate
+  recipe. Every weight it already held under the same rule -- the int8 trunk and lm_head, the E4M3
+  connection and MTP matrices, the ten bf16 experts -- was copied from it; the routed experts were
+  quantised by this run. Without `--reuse` the same command quantises those weights from the
+  checkpoint by the same rules.
+- `CALIB` is the calibration directory the GPTQ rule reads: each layer's routed-expert input Gram
+  from ten million tokens of the bf16 model, and each expert's own down-projection Gram from one
+  million, recorded with the engine's calibration mode on a mixed chat, tool-use and code corpus.
+  It is not published.
+- The published file is that output under this repository's name, with the local directory
+  removed from the `created by`, recipe and calibration-path strings in its header; the weights
+  are the converter's bytes.
+
+The recipe, as `rad-info --recipe qwen3.8-next-flash-fp8-iq4r-moe.rad` prints it; the first
+rule that matches a weight decides it, and everything no rule names is the checkpoint's bf16:
+
+```
+blk.34.ffn_*_exps.407.weight cast dtype=bf16
+blk.34.ffn_*_exps.496.weight cast dtype=bf16
+blk.44.ffn_*_exps.292.weight cast dtype=bf16
+blk.44.ffn_*_exps.350.weight cast dtype=bf16
+blk.46.ffn_*_exps.290.weight cast dtype=bf16
+blk.46.ffn_*_exps.392.weight cast dtype=bf16
+blk.47.ffn_*_exps.122.weight cast dtype=bf16
+blk.47.ffn_*_exps.143.weight cast dtype=bf16
+blk.47.ffn_*_exps.399.weight cast dtype=bf16
+blk.47.ffn_*_exps.445.weight cast dtype=bf16
+blk.*.ffn_*_exps.*.weight gptq table=w4nl group=64 scale=fp8_e4m3 scale2=f32 block2=*x* scale2_value=0.0001220703125 transform=fwht128 rule=search cd=3 calib=calib/w4nl-calib
+*_hc_down.weight rtn codes=fp8_e4m3 group=128 scale=f32
+*_hc_up.weight rtn codes=fp8_e4m3 group=80 scale=f32
+mtp.fc_hidden.weight rtn codes=fp8_e4m3 group=128 scale=f32
+mtp.fc_embedding.weight rtn codes=fp8_e4m3 group=128 scale=f32
+blk.*.attn_qg.weight rtn codes=i8 group=128 scale=bf16 clamp=sym rule=search
+blk.*.attn_k.weight rtn codes=i8 group=128 scale=bf16 clamp=sym rule=search
+blk.*.attn_v.weight rtn codes=i8 group=128 scale=bf16 clamp=sym rule=search
+blk.*.attn_output.weight rtn codes=i8 group=128 scale=bf16 clamp=sym rule=search
+blk.*.ssm_inz.weight rtn codes=i8 group=128 scale=bf16 clamp=sym rule=search
+blk.*.ssm_out.weight rtn codes=i8 group=128 scale=bf16 clamp=sym rule=search
+blk.*.ffn_gate_up_shexp.weight rtn codes=i8 group=128 scale=bf16 clamp=sym rule=search
+blk.*.ffn_down_shexp.weight rtn codes=i8 group=128 scale=bf16 clamp=sym rule=search
+output.weight rtn codes=i8 group=128 scale=bf16 clamp=sym rule=search
+blk.*.ple_ngram.weight rtn codes=fp8_e4m3 block=*x* scale=bf16 rule=fixed scale_value=1.99317932128906e-4
+mtp.draft_head.weight rtn codes=u2 zero=u8 group=128 scale=f16
+```
+
 ## License
 
 The Qwen community license of the base model; see `LICENSE`.