Add Q2_K/Q8_0 DSpark auxiliary

Deterministic q2_k-q8_0 build from pinned DeepSeek V4 Flash 0731 shards 46-48, with complete source verification, structural and numeric validation, and a byte-identical repeat build.

DeepSeek-V4-Flash-0731-DSpark-Drafter-Q2_K-Q8_0.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a3f1156e1e4d2f46e1d4367b1bd09e6923adee61aecd41e880e233df33d30a7b
+size 6971241728

README.md

@@ -29,20 +29,24 @@ pin, and build results are published alongside the artifact.
 | File | Bytes | SHA-256 |
 |---|---:|---|
 | `DeepSeek-V4-Flash-0731-DSpark-Drafter-MXFP4-Q8_0.gguf` | 10,897,110,272 | `5c8a1ce75e4797c170e22ac65080281a1b694c77f2e4fa7a703fbf09c5de0e33` |
+| `DeepSeek-V4-Flash-0731-DSpark-Drafter-Q2_K-Q8_0.gguf` | 6,971,241,728 | `a3f1156e1e4d2f46e1d4367b1bd09e6923adee61aecd41e880e233df33d30a7b` |
 
-The artifact uses GGUF v3 with the custom architecture
-`deepseek_v4_flash_dspark_draft` and contains 81 tensors:
+Both artifacts use GGUF v3 with the custom architecture
+`deepseek_v4_flash_dspark_draft` and contain the same 81 tensors. They differ
+only in the encoding of the nine routed-expert tensors:
 
 | Output type | Tensors | Transformation |
 |---|---:|---|
 | F32 | 45 | BF16 expanded exactly to F32, or source F32 bytes preserved |
 | F16 | 2 | BF16 Markov heads rounded to F16 |
 | Q8_0 | 25 | FP8 E4M3 weights with E8M0 block scales dequantized, then quantized to Q8_0 |
-| MXFP4 | 9 | Routed-expert FP4 nibbles and E8M0 scales repacked without requantization |
+| MXFP4 | 9 | Routed experts in the `mxfp4-q8_0` artifact: FP4 nibbles and E8M0 scales repacked without requantization |
+| Q2_K | 9 | Routed experts in the `q2_k-q8_0` artifact: source MXFP4 decoded, then quantized to Q2_K |
 
-The MXFP4 transformation is bit-preserving. Q8_0 and F16 transformations are
-lossy and are checked against explicit numeric error bounds. This repository
-does not claim that the complete conversion is lossless.
+The MXFP4 transformation is bit-preserving. Q2_K, Q8_0, and F16
+transformations are lossy and are checked against explicit numeric error
+bounds. This repository does not claim that either complete conversion is
+lossless.
 
 ## Pinned source
 
@@ -72,12 +76,14 @@ uv pip install --python .venv/bin/python \
 .venv/bin/python -m unittest discover -s tests -v
 .venv/bin/python scripts/reproduce.py \
   --sources sources \
-  --recipe mxfp4-q8_0 \
+  --recipe q2_k-q8_0 \
   --manifest-dir manifest \
   --repeat-check
 sha256sum -c SHA256SUMS
 ```
 
+Use `--recipe mxfp4-q8_0` to reproduce the MXFP4/Q8_0 artifact instead.
+
 `--repeat-check` performs a second clean conversion and requires its complete
 GGUF SHA-256 to match the verified first build byte for byte. The converter
 memory-maps the source shards and processes FP8 tensors in bounded bands. It
@@ -86,8 +92,9 @@ does not require PyTorch, CUDA, safetensors, a GGUF package, or a GPU.
 ## Verification
 
 The included verifier checks pinned source hashes and tensor inventories, GGUF
-structure and metadata, exact F32/F16 bytes, exact MXFP4 repacking, and Q8_0
-encoding plus reconstruction error against a fresh FP8 source decode.
+structure and metadata, exact F32/F16 bytes, exact MXFP4 repacking, Q2_K
+reconstruction error, and Q8_0 encoding plus reconstruction error against a
+fresh source decode.
 
 Reference-build results:
 
@@ -96,13 +103,18 @@ Reference-build results:
 | Tensor count | 81 |
 | Exact F32/F16 tensors | 47 |
 | Exact MXFP4 tensors | 9 |
+| Q2_K tensors | 9 |
+| Maximum Q2_K block-relative error | 0.49072265625 |
+| Q2_K limit | 0.5 |
 | Maximum Q8_0 block-relative error | 0.00396728515625 |
 | Q8_0 limit | 0.01 |
 | Repeat build | byte-identical |
 
-Machine-readable observations are in
-[`manifest/validation.json`](manifest/validation.json) and
-[`manifest/build.json`](manifest/build.json).
+Machine-readable observations are in the MXFP4/Q8_0
+[`validation`](manifest/validation.json) and [`build`](manifest/build.json)
+manifests, and the Q2_K/Q8_0
+[`validation`](manifest/validation-q2_k-q8_0.json) and
+[`build`](manifest/build-q2_k-q8_0.json) manifests.
 
 ## Compatibility and limitations
 

SHA256SUMS

@@ -1 +1,2 @@
 5c8a1ce75e4797c170e22ac65080281a1b694c77f2e4fa7a703fbf09c5de0e33  DeepSeek-V4-Flash-0731-DSpark-Drafter-MXFP4-Q8_0.gguf
+a3f1156e1e4d2f46e1d4367b1bd09e6923adee61aecd41e880e233df33d30a7b  DeepSeek-V4-Flash-0731-DSpark-Drafter-Q2_K-Q8_0.gguf

manifest/build-q2_k-q8_0.json

@@ -0,0 +1,38 @@
+{
+  "artifact": {
+    "filename": "DeepSeek-V4-Flash-0731-DSpark-Drafter-Q2_K-Q8_0.gguf",
+    "sha256": "a3f1156e1e4d2f46e1d4367b1bd09e6923adee61aecd41e880e233df33d30a7b",
+    "size": 6971241728
+  },
+  "build_inputs": {
+    "manifest/source.json": "bd7744f0efffd71d45e6233bfd8008e02acf422e01d701f82277d8821938b9c5",
+    "requirements-linux-aarch64-py314.lock": "f66db1728c9a0702b511a12bfe8cffbfab0fba4fbbf56f10d0e1d96ee70ee66d",
+    "scripts/convert.py": "319dc5083a5dc1b0a3cbce16702ae2378594b415ecf3f4198b2b84662fe31c43",
+    "scripts/download_sources.py": "2c39fff99a0c91b3ad099a5d685c394563e7b7a963b0a88cd1f668ee6bee51d1",
+    "scripts/formats.py": "4963ff62387bdbdbf0276b9db98e53da04ae621cb2cad231330cfff9f1533149",
+    "scripts/recipe.py": "661bf7b14ee5f2fc46770b5494cf6641407b5268c83ef5de6fc2d3e77e25b937",
+    "scripts/reproduce.py": "05956fbff7c6e32c9eeab6a959fc3d4dcd29d04a097296ea66cd000c64bfaa45",
+    "scripts/verify.py": "9073c95f157bee37c0fdea5d26b029b9495ef63f1c46e1d2022afdceb5cc9cc0"
+  },
+  "environment": {
+    "machine": "aarch64",
+    "numpy": "2.5.1",
+    "platform": "Linux-6.17.0-1026-nvidia-aarch64-with-glibc2.39",
+    "python": "3.14.6"
+  },
+  "format": {
+    "architecture": "deepseek_v4_flash_dspark_draft",
+    "gguf_version": 3,
+    "recipe": "q2_k-q8_0",
+    "recipe_version": 1
+  },
+  "repeat_build": {
+    "performed": true,
+    "sha256": "a3f1156e1e4d2f46e1d4367b1bd09e6923adee61aecd41e880e233df33d30a7b",
+    "status": "byte-identical"
+  },
+  "source": {
+    "repository": "deepseek-ai/DeepSeek-V4-Flash-0731",
+    "revision": "9e165c30e2704aec5d9d593cce3eebd58bbef1cb"
+  }
+}

manifest/validation-q2_k-q8_0.json

@@ -0,0 +1,48 @@
+{
+  "artifact": {
+    "filename": "DeepSeek-V4-Flash-0731-DSpark-Drafter-Q2_K-Q8_0.gguf",
+    "sha256": "a3f1156e1e4d2f46e1d4367b1bd09e6923adee61aecd41e880e233df33d30a7b",
+    "size": 6971241728
+  },
+  "gguf": {
+    "alignment": 32,
+    "data_start": 5824,
+    "tensor_count": 81,
+    "types": {
+      "F16": 2,
+      "F32": 45,
+      "Q2_K": 9,
+      "Q8_0": 25
+    },
+    "version": 3
+  },
+  "recipe": "q2_k-q8_0",
+  "source": {
+    "files": {
+      "model-00046-of-00048.safetensors": {
+        "sha256": "5db924ca907e0d93acd975bd5079c3662717f9ac709f23d079bd8f816d29d9dd",
+        "size": 3610455184
+      },
+      "model-00047-of-00048.safetensors": {
+        "sha256": "62816173f9f6e136b20b48e3b6f16613ac9ea02b5603f636928b253244a548bd",
+        "size": 3560111960
+      },
+      "model-00048-of-00048.safetensors": {
+        "sha256": "cc43742bd24ae6bcdea343a91442f6f66aed2cfebcc6b235470204851ce2f8a9",
+        "size": 3692775244
+      }
+    },
+    "repository": "deepseek-ai/DeepSeek-V4-Flash-0731",
+    "revision": "9e165c30e2704aec5d9d593cce3eebd58bbef1cb"
+  },
+  "validation": {
+    "f16_exact_tensors": 2,
+    "f32_exact_tensors": 45,
+    "q2_k_limit": 0.5,
+    "q2_k_max_block_relative_error": 0.49072265625,
+    "q2_k_tensors": 9,
+    "q8_0_limit": 0.01,
+    "q8_0_max_block_relative_error": 0.00396728515625,
+    "status": "pass"
+  }
+}