Skip to content

Qwen3.8-Flash-Next on any consumer hardware: one-click install for Windows / Linux. Strata inference engine, OpenAI/Anthropic API on localhost, optional image input.

C++

docs: DEEPSEEK_MAC.md with the numbers of the 0.2.7.1 build

Niko12215729a6e50a72
diff --git a/docs/DEEPSEEK_MAC.md b/docs/DEEPSEEK_MAC.md
index 6d140ea67..ee2ee6e62 100644
--- a/docs/DEEPSEEK_MAC.md
+++ b/docs/DEEPSEEK_MAC.md
@@ -1,21 +1,21 @@
# DeepSeek-V4.1-Flash on a Mac
-DeepSeek-V4.1-Flash runs on Apple Silicon Macs through its own Metal engine, `strata-dsv41`. It is a separate download
-from the Qwen engine: the app and `strata launch` fetch it the first time you pick DeepSeek. This page says what it
-needs, what it does on the machine it was measured on, and what it cannot do yet.
+DeepSeek-V4.1-Flash runs on Apple Silicon Macs on its own Metal engine, `strata-dsv41`. It is a separate download from the
+Qwen engine: the app and `strata launch` fetch it the first time you pick DeepSeek. This page lists what it needs, what it did
+on the one Mac it was measured on, and what it cannot do yet.
-> **Short version:** a Mac with 64 GB of memory or more, macOS 15 or newer, and about 260 GB of free disk space on a fast
-> drive. Text only (no pictures). The model does not fit in memory, so part of it is read from the SSD for every word.
-> On an M5 Max with 128 GB, DeepSeek 2.00 answers at 15-19 tokens a second and reads a 3,000-token prompt in under 8 seconds.
+> **Short version:** a Mac with 64 GB of memory or more, macOS 15 or newer, and 260 to 280 GB of free disk on a fast drive.
+> Text only, no pictures. The model does not fit in memory, so part of it is read from the SSD for every word. On an M5 Max with
+> 128 GB, DeepSeek 2.00 answers a short question at 15 to 18 tokens a second and reads a 3,000-token prompt in about 11 seconds.
## What it needs
| | |
|---|---|
| Mac | Apple Silicon (M1 or newer), macOS 15 or newer. Measured on an M5 Max with 128 GB. |
-| Memory | 64 GB or more is a good fit. Less still starts, but most of the model is read from the SSD for every word, so it is much slower; setup says so and asks before it goes on. |
-| Disk | 2.00 set: 257 GB. 2.25 set: 274 GB (the 111 GB n-gram table is shared). Put the model on the internal SSD or an NVMe drive, not a USB or network drive. |
-| Pictures | Off. DeepSeek's picture reader has not been run on the Metal engine, so a Mac reads text only (Qwen on the Mac keeps its pictures). |
+| Memory | 64 GB or more. A Mac with less still starts, but most of the model is read from the SSD for every word, so it is much slower. Setup says so and asks before it goes on. |
+| Disk | 2.00 set: 257 GB. 2.25 set: 274 GB (the 111 GB n-gram table is shared). Use the internal SSD or an NVMe drive, not a USB or network drive. |
+| Pictures | Off. DeepSeek's picture reader was not run on the Metal engine, so a Mac reads text only. Qwen on the Mac keeps its pictures. |
## Install it
@@ -28,43 +28,53 @@ strata launch --family deepseek41 --model ist-das-2.00bpw --plan # what it w
strata launch --family deepseek41 --model ist-das-2.00bpw
```
-Setup downloads the set from Hugging Face (pinned revision, size and SHA-256 checked, resumable), builds the pack (a few
-MB), downloads the DeepSeek engine package for macOS from the signed update feed, and runs a 16-word test question
-before it starts. Files you already have: `--gguf-dir DIR`.
+Setup downloads the set from Hugging Face (pinned revision, size and SHA-256 checked, resumable), builds the pack (a few MB),
+downloads the DeepSeek engine package for macOS from the signed update feed, and runs a short test question before it starts.
+Files you already have: `--model-path DIR`.
-## What it does (M5 Max, 128 GB, 2026-10-25 and -26)
+## What it did (M5 Max, 128 GB, 0.2.7.1 build)
+
+Speeds are from the packaged engine with the shipped defaults, one fresh process per run, two runs each.
| | 2.00 bpw | 2.25 bpw |
|---|---|---|
-| Answer speed, code prompt | 19.2 tokens/s | 14.4-15.8 |
-| Answer speed, agent prompt | 19.7 | 12.4-12.6 |
-| 3K prompt | 362-377 tokens/s | 380-427 |
-| 32K prompt | 303 | 306-404 |
-| Read from the SSD per word | 40 MB | 93 MB |
-| Quality against the original model (KL, 4,995 positions) | 0.1709 | 0.1102 |
-
-The 2.25 set is the better of the two (about 7% above the original model's perplexity, against about 15%), the 2.00 set is
-faster and smaller. Both numbers come from the engine's own scoring run with a fixed expert placement
+| Short answer, code prompt | 16.6 to 17.9 tokens/s | 13.4 to 15.0 |
+| Short answer, agent prompt | 15.1 to 16.6 | 13.1 to 14.9 |
+| After a 3,000-token prompt | 13.1 to 14.8 | 11.2 to 12.5 |
+| After a 14,000-token prompt | 14.9 to 15.4 | 10.3 to 12.1 |
+| 3,000-token prompt | 270 to 275 tokens/s | 275 to 308 |
+| 14,000-token prompt | 536 to 549 | 569 to 632 |
+| 32,000-token prompt | about 600 to 630 | about 720 |
+| Read from the SSD per word | 48 MB | 111 MB |
+| Quality against the original model (KL, 4,995 positions) | 0.170 | 0.111 |
+
+The 2.25 set is the better of the two (about 8% above the original model's perplexity on the Mac's scoring run, against about 17%). The 2.00 set is
+faster and smaller. The quality numbers come from the engine's scoring run with a fixed expert placement
(`--profile-readonly`); a run that learns the placement as it goes moves the last digits.
-How the memory is used: the engine keeps about 71 GB of the experts in memory on a 128 GB Mac (the rest of the memory is
-left to macOS and its file cache; a Mac with less memory keeps less) and reads the remaining experts from the SSD in place.
-A decode token is limited by those SSD reads and by a CPU lane that computes the experts the GPU does not hold, not by the
-GPU.
+How the memory is used: the engine keeps about 68 GB of the experts in memory on a 128 GB Mac. The rest of the memory goes to
+macOS and its file cache, and a Mac with less memory keeps fewer experts. The remaining experts are read from the SSD in place.
+A decode token is limited by those SSD reads and by a CPU lane that computes the experts the GPU does not hold, not by the GPU.
+If another program already holds a lot of memory (a browser, another model's server), the engine keeps fewer experts so that
+macOS does not run out. That costs some speed and keeps the Mac responsive.
+
+The needle test (a code hidden in a long text, asked for at the end) finds it at 14,600, 17,000, 19,500 and 26,200 tokens.
## Speed setting
-DeepSeek's Speed choice (Faster = 5 experts a word, Full quality = 6) works as on a PC. On a Mac the engine's own default
-is Full quality: that is what was measured and gated.
+DeepSeek's Speed choice (Faster = 5 experts a word, Full quality = 6) works as on a PC. On a Mac the engine's own default is
+Full quality. That is what was measured and scored.
## Limits
- Text only. A picture sent to a DeepSeek model on a Mac is refused.
-- The drafter (DSpark) is off: it was exact but not faster on the Mac.
-- Longer than 32K of context was not measured on a Mac.
-- Smaller Macs: nothing under 128 GB was run with DeepSeek. The plan screen shows how much of the model your Mac keeps in
- memory and warns when it is under a quarter of it; that estimate is the engine's own rule, not a measurement.
+- The drafter (DSpark) is off. It gave the same answers but no speed-up on the Mac.
+- Context longer than about 32,000 tokens was not measured on a Mac.
+- Smaller Macs: nothing under 128 GB was run with DeepSeek. The plan screen shows how much of the model your Mac keeps in memory
+ and warns when that is under a quarter of it. That estimate follows the engine's own rule and is not a measurement.
- The 1.75 bpw set shares the 2.00 set's kernel but was not run on a Mac, so it is not offered there.
+- Two runs of the same prompt can differ in the last digits of a score, because the engine moves experts between memory and the
+ SSD as it learns what the prompt needs. The answers stay equally good.
## Building and packaging the engine (developers)
@@ -73,10 +83,10 @@ bash packaging/mac/build_dsv41_engine.sh # configure, build, stage, p
```
It needs the Xcode Command Line Tools, cmake and ninja (a pip venv is enough), Apple's metal-cpp and the llama.cpp tree that
-setup pins (`METAL_CPP_DIR`, `STRATA_GGML_DIR`). The result is `pkg/engine/<version>-macos-arm64-metal/` with
-`strata-dsv41`, the Metal sources it compiles at start-up and `engine-component.json`; the package is unsigned, the feed key
-signs it (`packaging/UPDATES.md`). The Qwen Metal engine is not rebuilt by this: the DeepSeek runtime rules sit behind
-`STRATA_DS4_RUNTIME`, compiled only into `strata_ds4`.
+setup pins (`METAL_CPP_DIR`, `STRATA_GGML_DIR`). The result is `pkg/engine/<version>-macos-arm64-metal/` with `strata-dsv41`,
+the Metal sources it compiles at start-up and `engine-component.json`. The package is unsigned; the feed key signs it
+(`packaging/UPDATES.md`). The Qwen Metal engine is not rebuilt by this: the DeepSeek runtime rules sit behind
+`STRATA_DS4_RUNTIME` and are compiled only into `strata_ds4`.
Diagnostics, all off by default: `DS4_CPU_LOG=1` (per-token phase times), `DS4_PROF=1` (per-kernel GPU time),
`DS4_GPU_TOTAL=1`, `DS4_ENC_TIMES=1`, `DS4_TLOG=1`, `DS4_TRACE_DISP=1`.