Qwen3.8-27B on 1 DGX Spark

• NVFP4 weights
  The NVFP4 manifest file size is the current RadixArk artifact. It is the inclusive main ledger for this pin. The external drafter is not inside it.
  One Spark unified pool after load. Manifest bytes are not a resident measurement.
  21921697280 bytes (20.416 GiB). cluster total, not one rank.
  Per-rank storage: unknown. File ledger.
  Subcomponents can overlap this main-weight row. Do not add them into the checkpoint.
  Access: Resident matmul.
  The research brief does not tie this current manifest revision to the measured run. These file bytes are not resident memory and are not a fit.
• Routed experts
  There is no routed-expert stack on Qwen3.8-27B. The 6B active figure belongs to Qwen3.8-Flash-Next and is not a weight size for this model.
  Not present. Do not substitute a 6B active count for these weights.
  File size unknown.
  Per-rank storage: unknown. Not in this recipe.
  This row can overlap main weights. Do not add it again.
  Access: Not in this recipe.
  No routed-expert bytes exist on Qwen3.8-27B.
• Dense backbone
  Qwen3.8-27B is a dense vision-language checkpoint. Its language stack is resident matmul. This row overlaps the main file and is not a second copy.
  Inside the NVFP4 file, in the unified pool. 64 layers: 48 Gated DeltaNet plus 16 attention layers.
  File size unknown.
  Per-rank storage: unknown. Overlaps the file ledger.
  Subcomponents can overlap this main-weight row. Do not add them into the checkpoint.
  Access: Resident matmul.
  Dense bytes were not split out of the main file.
• Engram
  This Qwen profile has no Engram table. A PLE lookup, when this model has one, is a different component.
  Not present.
  File size unknown.
  Per-rank storage: unknown. Not in this recipe.
  This row can overlap main weights. Do not add it again.
  Access: Not in this recipe.
  No Engram bytes exist on this profile.
• PLE
  Qwen3.8-27B does not use a 51B n-gram PLE table. SSD offload for this model is not the Flash-Next file-backed PLE path.
  Not present.
  File size unknown.
  Per-rank storage: unknown. Not in this recipe.
  This row can overlap main weights. Do not add it again.
  Access: Not in this recipe.
  No PLE bytes exist on Qwen3.8-27B.
• Vision tower
  Qwen3.8-27B is a vision-language checkpoint. The vision tensors are inside the main NVFP4 file. This text C1 cell does not name a separate projector and does not say whether the vision tower was loaded. That is not the same as vision being absent, and it is not a second copy of the 4090 F16 projector.
  Inside the inclusive main ledger. Enabled or loaded residency for this cell is unknown.
  File size unknown.
  Per-rank storage: unknown. Overlaps the file ledger.
  Subcomponents can overlap this main-weight row. Do not add them into the checkpoint.
  Access: Resident matmul.
  Whether this text cell loaded the vision tower was not stated.
  Vision bytes were not split out of the main artifact.
• External DSpark
  RadixArk/Qwen3.8-27B-DSpark is a separate artifact with its own license field. It is not inside the NVFP4 file and it is not the internal MTP.
  Separate drafter file. License field other. Current manifest bytes are not the timed revision.
  3714723322 bytes (3.460 GiB). cluster total, not one rank.
  Per-rank storage: unknown. Separate from the file ledger.
  This row can overlap main weights. Do not add it again.
  Access: Resident matmul.
  The research brief does not tie this current manifest revision to the measured run. These file bytes are not resident memory and are not a fit.
  Do not add these bytes into the NVFP4 ledger and call the sum a fit.
• KV cache
  KV is stored for the 16 attention layers, not for all 64 layers. Recurrent state is the other row. A cookbook byte-per-token anchor is not multiplied into a fit here.
  KV dtype for this one-node cell was not printed. Attention cache only. Not a 64-layer cache.
  File size unknown.
  Per-rank storage: unknown. Runtime, not weights.
  This row can overlap main weights. Do not add it again.
  Access: Attention cache.
  This module does not multiply a per-token width by context.
  Native context 262144 and the 1M extension are not automatically verified by a shorter recipe.
• Recurrent state
  48 Gated DeltaNet layers keep a fixed recurrent state. That state is present at short context. It is not the KV cache, and a cookbook's approximate megabyte label is not stored as bytes.
  The Spark's one unified pool. Fixed state slots, separate from the 16 attention KV layers.
  File size unknown.
  Per-rank storage: unknown. Runtime, not weights.
  This row can overlap main weights. Do not add it again.
  Access: Recurrent state.
  Slot count depends on the runtime. One slot is not the whole state budget.
  Approximate megabytes were not converted.
• Workspace
  No process-memory integer was printed for this one-node cell.
  The same unified pool. Host RAM is not a second budget.
  File size unknown.
  Per-rank storage: unknown. Runtime, not weights.
  This row can overlap main weights. Do not add it again.
  Access: Runtime scratch.
  No workspace byte count was published.

• Pinned source Forum report cited by the research brief. The revision field is the image digest named there, not a git commit of the forum page. The page itself was not fetched into the cache. febfb971c7352570fc445c466ebd6ffc9d896024958e544a60f2137fd85856b1
• Pinned source Current NVFP4 manifest. Not tied to the timed run. 319f741cce68d7914884900c138a1fbb70a42f30
• Pinned source Current DSpark manifest, license field other. Not tied to the timed run. b9a5dbdf03bc999c6c73c426b19c2d9041cea393
• Pinned source Official BF16 scope only. 1d4bf0f2ff6012fd82039f2fa52739d0dd7c60c0

Speed figures in the copied brief are maintainer observations, not a measurement by this site.