Qwen2.5-7B-Instruct (Q4_K_M) for AMD Loom
A Loom package (.loom programs and a model.json card) for the unmodified GGUF checkpoint bartowski/Qwen2.5-7B-Instruct-GGUF (Qwen2.5-7B-Instruct-Q4_K_M.gguf), for the AMD Ryzen AI Max+ 395 "Strix Halo", AMD Radeon AI PRO R9700 and AMD Radeon gfx1201 16 GB class (RX 9070 XT, RX 9070) GPUs, generated with the loom-skills workflow. Learn more on the Loom ecosystem website.
Quick start
Open your GPU and copy its commands.
AMD Ryzen AI Max+ 395 "Strix Halo" (gfx1151)
pipx install "hrx-loom[hf]==0.2.0"
loom-fetch huggingface:loom-fun-staging/bartowski-Qwen2.5-7B-Instruct-loom:chat --materialize=copy --output=Qwen2.5-7B-Instruct
loom-serve --model=Qwen2.5-7B-Instruct '--device=amdgpu://?completion_spin_ns=4294967295' --pool_backing=fixed --parameter_mode=preload
--materialize=copyputs copies of the fetched package files in the output directory.completion_spin_ns=4294967295enables active waiting for GPU completions (uses CPU while GPU work is in flight); the card's performance measurements use this setting.--pool_backing=fixedcommits the preset's memory pool up front instead of growing elastic backing.--parameter_mode=preloadonly changes how the weights are loaded on the host, so the performance rows were measured without it unless their job served with it.
In another terminal, send a request:
curl http://127.0.0.1:8080/v1/chat/completions \
-H 'Content-Type: application/json' \
-d '{"model": "Qwen/Qwen2.5-7B-Instruct", "messages": [{"role": "user", "content": "Hello"}]}'
AMD Radeon AI PRO R9700 (gfx1201, 32 GB)
pipx install "hrx-loom[hf]==0.2.0"
loom-fetch huggingface:loom-fun-staging/bartowski-Qwen2.5-7B-Instruct-loom:chat --materialize=copy --output=Qwen2.5-7B-Instruct
loom-serve --model=Qwen2.5-7B-Instruct '--device=amdgpu://?completion_spin_ns=4294967295' --pool_backing=fixed --parameter_mode=preload
--materialize=copyputs copies of the fetched package files in the output directory.completion_spin_ns=4294967295enables active waiting for GPU completions (uses CPU while GPU work is in flight); the card's performance measurements use this setting.--pool_backing=fixedcommits the preset's memory pool up front instead of growing elastic backing.--parameter_mode=preloadonly changes how the weights are loaded on the host, so the performance rows were measured without it unless their job served with it.
In another terminal, send a request:
curl http://127.0.0.1:8080/v1/chat/completions \
-H 'Content-Type: application/json' \
-d '{"model": "Qwen/Qwen2.5-7B-Instruct", "messages": [{"role": "user", "content": "Hello"}]}'
Radeon gfx1201, 16 GB (RX 9070 XT, RX 9070)
pipx install "hrx-loom[hf]==0.2.0"
loom-fetch huggingface:loom-fun-staging/bartowski-Qwen2.5-7B-Instruct-loom:chat-16gb --materialize=copy --output=Qwen2.5-7B-Instruct
loom-serve --model=Qwen2.5-7B-Instruct '--device=amdgpu://?completion_spin_ns=4294967295' --memory_bytes=15000000000 --state_slab_bytes=4mib --weight_slab_bytes=64mib --parameter_mode=preload
--materialize=copyputs copies of the fetched package files in the output directory.completion_spin_ns=4294967295enables active waiting for GPU completions (uses CPU while GPU work is in flight); the card's performance measurements use this setting.--memory_bytessets the serving memory budget for this GPU.--state_slab_bytessets the allocation slab size for state memory.--weight_slab_bytessets the allocation slab size for weight memory.--parameter_mode=preloadonly changes how the weights are loaded on the host, so the performance rows were measured without it unless their job served with it.
In another terminal, send a request:
curl http://127.0.0.1:8080/v1/chat/completions \
-H 'Content-Type: application/json' \
-d '{"model": "Qwen/Qwen2.5-7B-Instruct", "messages": [{"role": "user", "content": "Hello"}]}'
The server listens on 127.0.0.1:8080 and answers OpenAI-style /v1/chat/completions requests.
Performance vs llama.cpp
Strix Halo (gfx1151)
| metric | llama.cpp | Loom | margin vs best llama.cpp | roofline | Loom % of roofline | llama.cpp Vulkan |
|---|---|---|---|---|---|---|
| decode tok/s, short prompt | 46.4 | 46.8 | +1 % | 49.4 | 95 % | 45.5 |
| decode tok/s, 2k context | 43.6 | 44.3 | +2 % | 48.2 | 92 % | 43.6 |
| decode tok/s, 32k context | 21.9 | 33.4 | +4 % | 34.8 | 96 % | 32.2 |
| prefill tok/s, 1k document | 1428.2 | 1968.4 | +38 % | 4452.4 | 44 % | 1136.4 |
| prefill tok/s, 2k document | 1434.2 | 2053.7 | +43 % | 4386.0 | 47 % | 1161.1 |
| prefill tok/s, 8k document | 1281.1 | 1864.2 | +46 % | 4025.6 | 46 % | 1014.7 |
| prefill tok/s, 16k document | 1085.3 | 1648.1 | +52 % | 3628.1 | 45 % | 861.7 |
| prefill tok/s, 32k document | 825.6 | 1204.9 | +46 % | 3029.8 | 40 % | 661.8 |
| 4 streams, aggregate decode tok/s | 144.8 | 175.4 | +21 % | 191.7 | 91 % | 143.6 |
| speculative decoding, drafted tokens accepted | n/a | n/a | n/a | n/a | n/a | n/a |
Methodology
- Compared: Loom against llama.cpp ROCm
llamacpp-rocm/b1374, llama.cpp Vulkanllamacpp-vulkan/b12211; the same GGUF, the same DevLab runner type, different jobs, the same prompts; Loom served with the active wait. Values are the median of 3 sessions (4 streams: median of 6); Loom is timed by the client, llama.cpp by its server. - llama.cpp measured on 2026-11-13 (DevLab Dispatch run 43392598780), Loom on 2026-11-17 (DevLab Dispatch run 44088665443): same runner type, different jobs (llama.cpp was not re-measured for this package).
- Out of the box on both engines: Loom serves the preset shipped in this package; llama.cpp runs its build defaults plus only the server arguments listed below (offload, context and slots). No tuning sweep on either engine.
- Drafting, 1-stream rows (decode, decode_2k, decode_32k, prefill): no drafting on either engine (no MTP head in the GGUF and no drafter published for this exact model). Each engine is loaded once for all these rows.
- 4 streams: both engines loaded with drafting off, 4 parallel requests, aggregate tokens per second.
- Rows: decode = short-context decode; decode_2k / decode_32k = decode after a ~2k / ~32k-token context; prefill_Nk = prompt processing of an N k-token prompt; streams4 = 4-stream aggregate decode. A margin within +-3 % (the measured same-runner noise) is a tie.
- llama.cpp ROCm server arguments:
1-stream rows: -ngl 99 -cram 0 -c 65536 -np 1; 4 streams: -ngl 99 -cram 0 -c 65536 -kvu -np 4. - llama.cpp Vulkan server arguments:
1-stream rows: -ngl 99 -cram 0 -c 36864 -np 1; 4 streams: -ngl 99 -cram 0 -c 147456 -np 4. - Measured on AMD DevLab Dispatch Strix Halo runners (
devlab-perfjobs); runners differ between sessions. - Margin vs best llama.cpp: (Loom / max(ROCm, Vulkan) - 1) Γ 100 %, using the available measured tokens/s for each row on this GPU. With only one valid backend, that backend is used; with neither, the margin is not measured. Raw backend values and their measurement sources are retained below and in the table.
- Measured 2026-11-17; device AMD Strix Halo (DevLab Dispatch runner X2-LA03-S01-D08), Linux; hrx-loom
0.2.0(hrx-system0cd370f, recovered); source: DevLab Dispatch run 44088665443. - llama.cpp Vulkan: measured 2026-11-13; source: DevLab Dispatch run 43392598780. Archive sha256
9759acd. - Roofline: the hardware ceiling for this GGUF on AMD Ryzen AI Max+ 395 "Strix Halo" (gfx1151) at 216 GB/s and 59 TFLOP/s, from loom-roofline
roofline.py(loom-skills1868f0b). - β not measured; n/a not applicable.
Radeon AI PRO R9700 (gfx1201, 32 GB)
| metric | llama.cpp | Loom | margin vs best llama.cpp | roofline | Loom % of roofline | llama.cpp Vulkan |
|---|---|---|---|---|---|---|
| decode tok/s, short prompt | 108.2 | 117.4 | +2 % | 140.3 | 84 % | 115.6 |
| decode tok/s, 2k context | 104.5 | 108.7 | -3 % | 136.7 | 80 % | 112.5 |
| decode tok/s, 32k context | 68.9 | 69.3 | -18 % | 98.8 | 70 % | 84.9 |
| prefill tok/s, 1k document | 4435.0 | 6040.3 | +36 % | 16149.4 | 37 % | 3191.0 |
| prefill tok/s, 2k document | 4449.4 | 6358.5 | +43 % | 15908.4 | 40 % | 3286.2 |
| prefill tok/s, 8k document | 3976.6 | 5869.9 | +48 % | 14601.3 | 40 % | 3009.3 |
| prefill tok/s, 16k document | 3422.0 | 5195.1 | +52 % | 13159.6 | 39 % | 2642.9 |
| prefill tok/s, 32k document | 2678.9 | 3824.5 | +43 % | 10989.5 | 35 % | 2113.5 |
| 4 streams, aggregate decode tok/s | 295.6 | 423.1 | +17 % | 543.9 | 78 % | 362.8 |
| speculative decoding, drafted tokens accepted | n/a | n/a | n/a | n/a | n/a | n/a |
Methodology
- Compared: Loom against llama.cpp ROCm
llamacpp-rocm/b1375, llama.cpp Vulkanllamacpp-vulkan/b12211; the same GGUF, the same DevLab runner type, different jobs, the same prompts; Loom served with the active wait. Values are the median of 3 sessions (4 streams: median of 6); Loom is timed by the client, llama.cpp by its server. - llama.cpp measured on 2026-11-14 (DevLab Dispatch run 43633449157), Loom on 2026-11-17 (DevLab Dispatch run 43993223590): same runner type, different jobs (llama.cpp was not re-measured for this package).
- Out of the box on both engines: Loom serves the preset shipped in this package; llama.cpp runs its build defaults plus only the server arguments listed below (offload, context and slots). No tuning sweep on either engine.
- Drafting, 1-stream rows (decode, decode_2k, decode_32k, prefill): no drafting on either engine (no MTP head in the GGUF and no drafter published for this exact model). Each engine is loaded once for all these rows.
- 4 streams: both engines loaded with drafting off, 4 parallel requests, aggregate tokens per second.
- Rows: decode = short-context decode; decode_2k / decode_32k = decode after a ~2k / ~32k-token context; prefill_Nk = prompt processing of an N k-token prompt; streams4 = 4-stream aggregate decode. A margin within +-3 % (the measured same-runner noise) is a tie.
- llama.cpp ROCm server arguments:
-ngl 99 -cram 0 -c 65536 -np 1. - llama.cpp Vulkan server arguments:
-ngl 99 -cram 0 -c 36864 -np 1. - Measured on AMD DevLab Dispatch R9700 runners (
devlab-perfjobs); runners differ between sessions. - Margin vs best llama.cpp: (Loom / max(ROCm, Vulkan) - 1) Γ 100 %, using the available measured tokens/s for each row on this GPU. With only one valid backend, that backend is used; with neither, the margin is not measured. Raw backend values and their measurement sources are retained below and in the table.
- Measured 2026-11-17; device DevLab Dispatch runner ephemeral-devlab-sm-01-02-r9700s-3-133738799824, AMD Radeon AI PRO R9700S (gfx1201), Linux; hrx-loom
0.2.0(hrx-system commit unknown, recovered); source: DevLab Dispatch run 43993223590. - llama.cpp Vulkan: measured 2026-11-14; source: DevLab Dispatch run 43633449157. Archive sha256
9759acd. - Roofline: the hardware ceiling for this GGUF on AMD Radeon AI PRO R9700S (gfx1201, 32 GB, 2.35 GHz) at 613 GB/s and 214 TFLOP/s, from loom-roofline
roofline.py(loom-skills203cf33). - β not measured; n/a not applicable.
Radeon RX 9070 XT
| metric | llama.cpp | Loom | margin vs best llama.cpp | roofline | Loom % of roofline | llama.cpp Vulkan |
|---|---|---|---|---|---|---|
| decode tok/s, short prompt | 109.2 | 117.7 | +0 % | 146.4 | 80 % | 117.5 |
| decode tok/s, 2k context | 105.1 | 107.8 | -6 % | 142.7 | 76 % | 114.1 |
| decode tok/s, 32k context | 70.8 | 69.3 | -19 % | 103.1 | 67 % | 85.9 |
| prefill tok/s, 1k document | 4346.5 | 6025.6 | +39 % | 14685.4 | 41 % | 3789.5 |
| prefill tok/s, 2k document | 4363.8 | 6388.3 | +46 % | 14466.3 | 44 % | 3868.3 |
| prefill tok/s, 8k document | 3898.2 | 5986.7 | +54 % | 13277.6 | 45 % | 3543.5 |
| prefill tok/s, 16k document | 3331.9 | 5294.5 | +59 % | 11966.7 | 44 % | 3114.8 |
| prefill tok/s, 32k document | 2620.7 | 3816.5 | +46 % | 9993.3 | 38 % | 2504.9 |
| 4 streams, aggregate decode tok/s | 301.1 | 429.8 | +13 % | 567.9 | 76 % | 379.1 |
| speculative decoding, drafted tokens accepted | n/a | n/a | n/a | n/a | n/a | n/a |
Methodology
- Compared: Loom against llama.cpp ROCm
llamacpp-rocm/b1375, llama.cpp Vulkanllamacpp-vulkan/b12211; the same GGUF, the same machine and session, the same prompts; Loom served with the active wait. Values are the median of 3 sessions (4 streams: median of 3); Loom is timed by the client, llama.cpp by its server. - Out of the box on both engines: Loom serves the preset shipped in this package; llama.cpp runs its build defaults plus only the server arguments listed below (offload, context and slots). No tuning sweep on either engine.
- Drafting, 1-stream rows (decode, decode_2k, decode_32k, prefill): no drafting on either engine (no MTP head in the GGUF and no drafter published for this exact model). Each engine is loaded once for all these rows.
- 4 streams: both engines loaded with drafting off, 4 parallel requests, aggregate tokens per second.
- Rows: decode = short-context decode; decode_2k / decode_32k = decode after a ~2k / ~32k-token context; prefill_Nk = prompt processing of an N k-token prompt; streams4 = 4-stream aggregate decode. A margin within +-3 % (the measured same-runner noise) is a tie.
- llama.cpp ROCm server arguments:
-ngl 99 -cram 0 -c 65536 -np 1. - llama.cpp Vulkan server arguments:
-ngl 99 -cram 0 -c 36864 -np 1. - Measured on AMD Ryzen 7 9800X3D 8-Core Processor, AMD Radeon RX 9070 XT, Ubuntu 26.04, a named machine (this GPU has no DevLab runner); both engines ran on it.
- Margin vs best llama.cpp: (Loom / max(ROCm, Vulkan) - 1) Γ 100 %, using the available measured tokens/s for each row on this GPU. With only one valid backend, that backend is used; with neither, the margin is not measured. Raw backend values and their measurement sources are retained below and in the table.
- Measured 2026-11-17; device AMD Ryzen 7 9800X3D 8-Core Processor, AMD Radeon RX 9070 XT, Ubuntu 26.04; hrx-loom
0.2.0(hrx-system0cd370f); source: named-machine run. - llama.cpp Vulkan: measured 2026-11-17; source: named-machine run. Archive sha256
9759acd. - Roofline: the hardware ceiling for this GGUF on AMD Radeon RX 9070 XT (gfx1201, 16 GB) at 640 GB/s and 195 TFLOP/s, from loom-roofline
roofline.py(loom-skills963ead9). - β not measured; n/a not applicable.
Features
| feature | status | detail |
|---|---|---|
| API | β | OpenAI-style /v1/chat/completions |
| tool calling | β | OpenAI-style tools and tool_calls |
| reasoning / thinking | β | not applicable |
| reasoning effort | β | not applicable |
| speculative decoding | β | not applicable |
| image input | β | not applicable |
| audio input / output | β | not applicable |
| sampling | β | temperature, top_p, top_k, seed |
| context size | β | 32,768 tokens per request |
| concurrency | β | 4 concurrent requests |
| max output tokens | β | 4,096 (preset max_tokens) |
Correctness
| gate | description | Strix Halo | Radeon AI PRO R9700 | Radeon RX 9070 XT |
|---|---|---|---|---|
kernel_tests |
Unit tests of the generated kernels against reference implementations | β pass (59/59) ΒΉ | β pass (59/59) Β² | β pass (59/59) Β³ |
golden |
llama.cpp reference recorded on the same GGUF: greedy tokens and top-5 log-probs, short prompts and 1k-32k documents | β pass β΄ | β pass β΅ | β pass βΆ |
compare |
Greedy output vs the llama.cpp reference, token by token; every divergence is a near-tie | β pass β΄ | β pass β΅ | β pass βΆ |
vision |
Image input vs llama.cpp with the same projector: identical prompt tokens, replies equal up to near-ties | β not applicable | β not applicable | β not applicable |
consistency |
Concurrent requests give the same output as the same requests run alone | β pass β΄ | β pass β΅ | β pass βΆ |
spec |
Speculative decoding on gives exactly the output of speculative decoding off | β not applicable | β not applicable | β not applicable |
agent |
Multi-turn tool-calling loop: exact prompt rendering, history reuse, valid tool calls, output vs llama.cpp | β pass β΄ | β pass β΅ | β pass βΆ |
agent_think_off |
The tool-calling loop with thinking off | β not applicable | β not applicable | β not applicable |
sampling |
temperature, top_k and seed behave as specified; first-token distribution matches llama.cpp | β pass β΄ | β pass β΅ | β pass βΆ |
logits |
Host logits readback (llm.logits on): at T=0 every returned row's argmax is the device's token, the top-k matches llama.cpp, greedy text is unchanged | ? not re-run on this package (the last pass was on a previous build, sha256 b989b3f; the nightly gates re-run it) |
? not re-run on this package (the last pass was on a previous build, sha256 b989b3f; the nightly gates re-run it) |
β no evidence |
decision |
Decision answers (choice, score, noul probabilities) vs llama.cpp on the same prompts | β not applicable | β not applicable | β not applicable |
- ΒΉ DevLab Dispatch run 44088665443, 2026-11-17; device AMD Strix Halo (DevLab Dispatch runner X2-LA03-S01-D08), Linux; hrx-loom
0.2.0(hrx-system0cd370f, recovered). - Β² DevLab Dispatch run 43993223590, 2026-11-17; device DevLab Dispatch runner ephemeral-devlab-sm-01-02-r9700s-3-133738799824, AMD Radeon AI PRO R9700S (gfx1201), Linux; hrx-loom
0.2.0(hrx-system commit unknown, recovered). - Β³ named-machine run, 2026-11-17; device AMD Ryzen 7 9800X3D 8-Core Processor, AMD Radeon RX 9070 XT, Ubuntu 26.04; hrx-loom
0.2.0(hrx-system0cd370f). - β΄ DevLab Dispatch run 44088665443, LEVEL=dev, 2026-11-17; device AMD Strix Halo (DevLab Dispatch runner X2-LA03-S01-D08), Linux; hrx-loom
0.2.0(hrx-system0cd370f, recovered). - β΅ DevLab Dispatch run 43993223590, LEVEL=dev, 2026-11-17; device DevLab Dispatch runner ephemeral-devlab-sm-01-02-r9700s-3-133738799824, AMD Radeon AI PRO R9700S (gfx1201), Linux; hrx-loom
0.2.0(hrx-system commit unknown, recovered). - βΆ named-machine run, LEVEL=dev, 2026-11-17; device AMD Ryzen 7 9800X3D 8-Core Processor, AMD Radeon RX 9070 XT, Ubuntu 26.04; hrx-loom
0.2.0(hrx-system0cd370f). - Gates are the loom-benchmark correctness gates (
gates.sh); a gate without a recorded run shows no evidence.
Files
model.json/model.json.lock: the card and the pinned source files (sha256 checked byloom-fetch).program/*.loom,weights.loom: the generated Loom programs.tokenizer.json: the tokenizer.llm_config.json: the configuration the programs were generated from.info.json: the measurements, gate evidence and versions this card shows; everything else comes from the files above.REPORT.md: how the values above were measured: setup, commands and the value of every run.
Versions: card schema loom-card/3; loom-skills unknown; hrx-loom unknown; card generator b931191.
Source files, fetched by loom-fetch from their own repositories:
- checkpoint: bartowski/Qwen2.5-7B-Instruct-GGUF @
8911e8a,Qwen2.5-7B-Instruct-Q4_K_M.gguf
Model tree for loom-fun-staging/bartowski-Qwen2.5-7B-Instruct-loom
Base model
Qwen/Qwen2.5-7B