Qwen3.6 35B-A3B
Qwen/Qwen3.6-35B-A3B
Qwen · 9 recorded configurations
Measured setup
Hardware
Runtime and tool
Quantisation
MTP draft depth
Every option opens a recorded setup. Where the corpus recorded nothing at this page's other settings, the option names the fields it switches with it. Allocation and runtime detail stay in the technical record below.
1× RTX 3090 · llama.cpp
UD-IQ4_XS · MTP off
PP · left axisTG · right axis
View exact measurements
| Context depth (tokens) | Fresh prefill PP (tok/s) | TG (tok/s) |
|---|---|---|
| 0 | 3,247.24 | 138.25 |
| 2,048 | 2,638.22 | 135.51 |
| 4,096 | 2,602.20 | 132.74 |
| 8,192 | 2,569.50 | 128.03 |
| 16,384 | 2,423.59 | 118.77 |
| 32,768 | 2,132.32 | 104.80 |
| 65,536 | 1,754.23 | 83.69 |
| 131,072 | 1,276.52 | 60.04 |
| 261,312 | 829.65 | 38.05 |
Concurrent-stream measurements
| Streams | Context depth (tokens) | Aggregate PP (tok/s) | Aggregate TG (tok/s) | TG / stream (tok/s) | Aggregate all (tok/s) |
|---|---|---|---|---|---|
| 2 | 8,192 | 2,250.17 | 192.03 | 96.02 | 496.20 |
| 3 | 8,192 | 1,938.66 | 209.01 | 69.67 | 549.09 |
| 4 | 8,192 | 1,863.74 | 213.00 | 53.25 | 576.35 |
| 5 | 8,192 | 1,742.51 | 217.17 | 43.43 | 598.13 |
| 6 | 8,192 | 1,771.55 | 222.09 | 37.01 | 619.81 |
| 7 | 8,192 | 1,752.86 | 220.03 | 31.43 | 620.37 |
| 8 | 8,192 | 1,716.00 | 226.41 | 28.30 | 641.51 |
Technical record and provenance
Source 1 · 2026-10-15
- Source run ID
- pub-20261016-trpro-qwen36-iq4-xs-context-2026-10-14-single-gpu-isolated-off-c1
- Page entry ID
- qwen3-6-35b-a3b-ud-iq4-xs-single-gpu-20261016-off-0f6923f607
- Entry sweep
- context
- Raw entry label
- Qwen3.6 35B-A3B UD-IQ4_XS — Single GPU — no splitting
- Configuration label
- Single GPU — no splitting
- Measured date
- 2026-10-15
- Hardware
- 1× NVIDIA GeForce RTX 3090
- Runtime
- llamaCpp · version: 0.3.0-dev (build 50, commit fc6545d) built with GNU 15.2.0 for Linux x86_64
- Weight placement
- gpu-resident
- Observed devices
- CUDA0
- Requested devices
- CUDA0
- Split mode
- Single GPU — no splitting
- Placement detail
- Single GPU — no splitting
- KV-cache precision
- Q8_0
- Flash Attention
- Enabled
- Requested context per slot
- 262,144 tokens
- Actual context per slot
- 262,144 tokens
- Native model context
- 262,144 tokens
- Batch size
- 512
- Microbatch size
- 512
- Prompt tokens
- 512
- Generated tokens
- 256
- Speculative decoding
- None (standard decoding)
- Warm-ups
- 1
- Measured repetitions
- 3
- Timing method
- server durations aligned to client completion timestamps; aggregate-all uses client wall time
- Throughput spreads
- KV 0, 1 stream: PP mean 3,247.24 t/s (min 3,207.76, max 3,276.72, SD 29.03, n=3); TG mean 138.25 t/s (min 137.95, max 138.79, SD 0.39, n=3); All mean 375.94 t/s (min 375.29, max 376.72, SD 0.59, n=3) · KV 2,048, 1 stream: PP mean 2,638.22 t/s (min 2,570.38, max 2,684.76, SD 49.06, n=3); TG mean 135.51 t/s (min 135.21, max 135.80, SD 0.24, n=3); All mean 362.88 t/s (min 361.36, max 363.92, SD 1.10, n=3) · KV 4,096, 1 stream: PP mean 2,602.20 t/s (min 2,590.94, max 2,619.21, SD 12.23, n=3); TG mean 132.74 t/s (min 132.68, max 132.79, SD 0.04, n=3); All mean 355.67 t/s (min 355.63, max 355.73, SD 0.04, n=3) · KV 8,192, 1 stream: PP mean 2,569.50 t/s (min 2,561.42, max 2,573.99, SD 5.73, n=3); TG mean 128.03 t/s (min 127.73, max 128.29, SD 0.23, n=3); All mean 343.87 t/s (min 343.08, max 344.48, SD 0.58, n=3) · KV 16,384, 1 stream: PP mean 2,423.59 t/s (min 2,408.46, max 2,439.38, SD 12.63, n=3); TG mean 118.77 t/s (min 118.70, max 118.89, SD 0.09, n=3); All mean 319.47 t/s (min 319.38, max 319.61, SD 0.10, n=3) · KV 32,768, 1 stream: PP mean 2,132.32 t/s (min 2,098.66, max 2,161.73, SD 25.92, n=3); TG mean 104.80 t/s (min 104.62, max 105.04, SD 0.18, n=3); All mean 281.79 t/s (min 281.22, max 282.83, SD 0.74, n=3) · KV 65,536, 1 stream: PP mean 1,754.23 t/s (min 1,749.42, max 1,756.96, SD 3.41, n=3); TG mean 83.69 t/s (min 83.50, max 83.87, SD 0.15, n=3); All mean 225.56 t/s (min 225.08, max 226.00, SD 0.38, n=3) · KV 131,072, 1 stream: PP mean 1,276.52 t/s (min 1,265.00, max 1,284.50, SD 8.34, n=3); TG mean 60.04 t/s (min 59.95, max 60.09, SD 0.06, n=3); All mean 162.11 t/s (min 161.79, max 162.31, SD 0.23, n=3) · KV 261,312, 1 stream: PP mean 829.65 t/s (min 828.80, max 830.30, SD 0.63, n=3); TG mean 38.05 t/s (min 38.01, max 38.11, SD 0.04, n=3); All mean 103.02 t/s (min 102.90, max 103.13, SD 0.09, n=3)
- Measurement outcomes
- 0 tokens · 1 stream: measured; 2,048 tokens · 1 stream: measured; 4,096 tokens · 1 stream: measured; 8,192 tokens · 1 stream: measured; 16,384 tokens · 1 stream: measured; 32,768 tokens · 1 stream: measured; 65,536 tokens · 1 stream: measured; 131,072 tokens · 1 stream: measured; 261,312 tokens · 1 stream: measured
- Notes
- Automated llama-server KV-depth sweep (off, concurrency=1). Existing depth is contextTokens; PP=512, TG=256, warmups=1, repetitions=3, Q8_0 K/V. Arithmetic mean and population standard deviation of measured repetitions. PP/TG server durations are aligned to client completion timestamps for concurrent phase spans; aggregate-all includes SSH/HTTP wall time. Synthetic repeated-token prompts. Engine was reused as installed; no build was performed. Publication projection of reviewed source measurements; values and raw evidence unchanged. Single GPU — no splitting.
- Source article
- /benchmarks/
- Raw log path
research/benchmarks/logs/trpro/2026-10-14-approved-models-single-gpu/qwen36-iq4-xs-context-isolated
Source 2 · 2026-10-15
- Source run ID
- pub-20261016-trpro-qwen36-iq4-xs-concurrency-2026-10-14-single-gpu-isolated-off-c2
- Page entry ID
- qwen3-6-35b-a3b-ud-iq4-xs-single-gpu-20261016-off-0f6923f607
- Entry sweep
- concurrency
- Raw entry label
- Qwen3.6 35B-A3B UD-IQ4_XS — Single GPU — no splitting
- Configuration label
- Single GPU — no splitting
- Measured date
- 2026-10-15
- Hardware
- 1× NVIDIA GeForce RTX 3090
- Runtime
- llamaCpp · version: 0.3.0-dev (build 50, commit fc6545d) built with GNU 15.2.0 for Linux x86_64
- Weight placement
- gpu-resident
- Observed devices
- CUDA0
- Requested devices
- CUDA0
- Split mode
- Single GPU — no splitting
- Placement detail
- Single GPU — no splitting
- KV-cache precision
- Q8_0
- Flash Attention
- Enabled
- Requested context per slot
- 9,024 tokens
- Actual context per slot
- 9,216 tokens
- Native model context
- 262,144 tokens
- Batch size
- 512
- Microbatch size
- 512
- Prompt tokens
- 512
- Generated tokens
- 256
- Speculative decoding
- None (standard decoding)
- Warm-ups
- 1
- Measured repetitions
- 3
- Timing method
- server durations aligned to client completion timestamps; aggregate-all uses client wall time
- Throughput spreads
- KV 8,192, 2 streams: PP mean 2,250.17 t/s (min 2,081.57, max 2,341.32, SD 119.35, n=3); TG mean 192.03 t/s (min 191.47, max 192.60, SD 0.46, n=3); All mean 496.20 t/s (min 489.15, max 500.25, SD 5.00, n=3)
- Measurement outcomes
- 8,192 tokens · 2 streams: measured
- Notes
- Automated llama-server KV-depth sweep (off, concurrency=2). Existing depth is contextTokens; PP=512, TG=256, warmups=1, repetitions=3, Q8_0 K/V. Arithmetic mean and population standard deviation of measured repetitions. PP/TG server durations are aligned to client completion timestamps for concurrent phase spans; aggregate-all includes SSH/HTTP wall time. Synthetic repeated-token prompts. Engine was reused as installed; no build was performed. Publication projection of reviewed source measurements; values and raw evidence unchanged. Single GPU — no splitting.
- Source article
- /benchmarks/
- Raw log path
research/benchmarks/logs/trpro/2026-10-14-approved-models-single-gpu/qwen36-iq4-xs-concurrency-isolated
Source 3 · 2026-10-15
- Source run ID
- pub-20261016-trpro-qwen36-iq4-xs-concurrency-2026-10-14-single-gpu-isolated-off-c3
- Page entry ID
- qwen3-6-35b-a3b-ud-iq4-xs-single-gpu-20261016-off-0f6923f607
- Entry sweep
- concurrency
- Raw entry label
- Qwen3.6 35B-A3B UD-IQ4_XS — Single GPU — no splitting
- Configuration label
- Single GPU — no splitting
- Measured date
- 2026-10-15
- Hardware
- 1× NVIDIA GeForce RTX 3090
- Runtime
- llamaCpp · version: 0.3.0-dev (build 50, commit fc6545d) built with GNU 15.2.0 for Linux x86_64
- Weight placement
- gpu-resident
- Observed devices
- CUDA0
- Requested devices
- CUDA0
- Split mode
- Single GPU — no splitting
- Placement detail
- Single GPU — no splitting
- KV-cache precision
- Q8_0
- Flash Attention
- Enabled
- Requested context per slot
- 9,024 tokens
- Actual context per slot
- 9,216 tokens
- Native model context
- 262,144 tokens
- Batch size
- 512
- Microbatch size
- 512
- Prompt tokens
- 512
- Generated tokens
- 256
- Speculative decoding
- None (standard decoding)
- Warm-ups
- 1
- Measured repetitions
- 3
- Timing method
- server durations aligned to client completion timestamps; aggregate-all uses client wall time
- Throughput spreads
- KV 8,192, 3 streams: PP mean 1,938.66 t/s (min 1,868.12, max 1,975.17, SD 49.89, n=3); TG mean 209.01 t/s (min 207.62, max 209.76, SD 0.99, n=3); All mean 549.09 t/s (min 543.55, max 552.12, SD 3.92, n=3)
- Measurement outcomes
- 8,192 tokens · 3 streams: measured
- Notes
- Automated llama-server KV-depth sweep (off, concurrency=3). Existing depth is contextTokens; PP=512, TG=256, warmups=1, repetitions=3, Q8_0 K/V. Arithmetic mean and population standard deviation of measured repetitions. PP/TG server durations are aligned to client completion timestamps for concurrent phase spans; aggregate-all includes SSH/HTTP wall time. Synthetic repeated-token prompts. Engine was reused as installed; no build was performed. Publication projection of reviewed source measurements; values and raw evidence unchanged. Single GPU — no splitting.
- Source article
- /benchmarks/
- Raw log path
research/benchmarks/logs/trpro/2026-10-14-approved-models-single-gpu/qwen36-iq4-xs-concurrency-isolated
Source 4 · 2026-10-15
- Source run ID
- pub-20261016-trpro-qwen36-iq4-xs-concurrency-2026-10-14-single-gpu-isolated-off-c4
- Page entry ID
- qwen3-6-35b-a3b-ud-iq4-xs-single-gpu-20261016-off-0f6923f607
- Entry sweep
- concurrency
- Raw entry label
- Qwen3.6 35B-A3B UD-IQ4_XS — Single GPU — no splitting
- Configuration label
- Single GPU — no splitting
- Measured date
- 2026-10-15
- Hardware
- 1× NVIDIA GeForce RTX 3090
- Runtime
- llamaCpp · version: 0.3.0-dev (build 50, commit fc6545d) built with GNU 15.2.0 for Linux x86_64
- Weight placement
- gpu-resident
- Observed devices
- CUDA0
- Requested devices
- CUDA0
- Split mode
- Single GPU — no splitting
- Placement detail
- Single GPU — no splitting
- KV-cache precision
- Q8_0
- Flash Attention
- Enabled
- Requested context per slot
- 9,024 tokens
- Actual context per slot
- 9,216 tokens
- Native model context
- 262,144 tokens
- Batch size
- 512
- Microbatch size
- 512
- Prompt tokens
- 512
- Generated tokens
- 256
- Speculative decoding
- None (standard decoding)
- Warm-ups
- 1
- Measured repetitions
- 3
- Timing method
- server durations aligned to client completion timestamps; aggregate-all uses client wall time
- Throughput spreads
- KV 8,192, 4 streams: PP mean 1,863.74 t/s (min 1,814.85, max 1,891.29, SD 34.66, n=3); TG mean 213.00 t/s (min 211.57, max 213.73, SD 1.01, n=3); All mean 576.35 t/s (min 571.43, max 578.93, SD 3.48, n=3)
- Measurement outcomes
- 8,192 tokens · 4 streams: measured
- Notes
- Automated llama-server KV-depth sweep (off, concurrency=4). Existing depth is contextTokens; PP=512, TG=256, warmups=1, repetitions=3, Q8_0 K/V. Arithmetic mean and population standard deviation of measured repetitions. PP/TG server durations are aligned to client completion timestamps for concurrent phase spans; aggregate-all includes SSH/HTTP wall time. Synthetic repeated-token prompts. Engine was reused as installed; no build was performed. Publication projection of reviewed source measurements; values and raw evidence unchanged. Single GPU — no splitting.
- Source article
- /benchmarks/
- Raw log path
research/benchmarks/logs/trpro/2026-10-14-approved-models-single-gpu/qwen36-iq4-xs-concurrency-isolated
Source 5 · 2026-10-15
- Source run ID
- pub-20261016-trpro-qwen36-iq4-xs-concurrency-2026-10-14-single-gpu-isolated-off-c5
- Page entry ID
- qwen3-6-35b-a3b-ud-iq4-xs-single-gpu-20261016-off-0f6923f607
- Entry sweep
- concurrency
- Raw entry label
- Qwen3.6 35B-A3B UD-IQ4_XS — Single GPU — no splitting
- Configuration label
- Single GPU — no splitting
- Measured date
- 2026-10-15
- Hardware
- 1× NVIDIA GeForce RTX 3090
- Runtime
- llamaCpp · version: 0.3.0-dev (build 50, commit fc6545d) built with GNU 15.2.0 for Linux x86_64
- Weight placement
- gpu-resident
- Observed devices
- CUDA0
- Requested devices
- CUDA0
- Split mode
- Single GPU — no splitting
- Placement detail
- Single GPU — no splitting
- KV-cache precision
- Q8_0
- Flash Attention
- Enabled
- Requested context per slot
- 9,024 tokens
- Actual context per slot
- 9,216 tokens
- Native model context
- 262,144 tokens
- Batch size
- 512
- Microbatch size
- 512
- Prompt tokens
- 512
- Generated tokens
- 256
- Speculative decoding
- None (standard decoding)
- Warm-ups
- 1
- Measured repetitions
- 3
- Timing method
- server durations aligned to client completion timestamps; aggregate-all uses client wall time
- Throughput spreads
- KV 8,192, 5 streams: PP mean 1,742.51 t/s (min 1,709.65, max 1,781.38, SD 29.59, n=3); TG mean 217.17 t/s (min 216.42, max 217.80, SD 0.57, n=3); All mean 598.13 t/s (min 595.46, max 601.00, SD 2.27, n=3)
- Measurement outcomes
- 8,192 tokens · 5 streams: measured
- Notes
- Automated llama-server KV-depth sweep (off, concurrency=5). Existing depth is contextTokens; PP=512, TG=256, warmups=1, repetitions=3, Q8_0 K/V. Arithmetic mean and population standard deviation of measured repetitions. PP/TG server durations are aligned to client completion timestamps for concurrent phase spans; aggregate-all includes SSH/HTTP wall time. Synthetic repeated-token prompts. Engine was reused as installed; no build was performed. Publication projection of reviewed source measurements; values and raw evidence unchanged. Single GPU — no splitting.
- Source article
- /benchmarks/
- Raw log path
research/benchmarks/logs/trpro/2026-10-14-approved-models-single-gpu/qwen36-iq4-xs-concurrency-isolated
Source 6 · 2026-10-15
- Source run ID
- pub-20261016-trpro-qwen36-iq4-xs-concurrency-2026-10-14-single-gpu-isolated-off-c6
- Page entry ID
- qwen3-6-35b-a3b-ud-iq4-xs-single-gpu-20261016-off-0f6923f607
- Entry sweep
- concurrency
- Raw entry label
- Qwen3.6 35B-A3B UD-IQ4_XS — Single GPU — no splitting
- Configuration label
- Single GPU — no splitting
- Measured date
- 2026-10-15
- Hardware
- 1× NVIDIA GeForce RTX 3090
- Runtime
- llamaCpp · version: 0.3.0-dev (build 50, commit fc6545d) built with GNU 15.2.0 for Linux x86_64
- Weight placement
- gpu-resident
- Observed devices
- CUDA0
- Requested devices
- CUDA0
- Split mode
- Single GPU — no splitting
- Placement detail
- Single GPU — no splitting
- KV-cache precision
- Q8_0
- Flash Attention
- Enabled
- Requested context per slot
- 9,024 tokens
- Actual context per slot
- 9,216 tokens
- Native model context
- 262,144 tokens
- Batch size
- 512
- Microbatch size
- 512
- Prompt tokens
- 512
- Generated tokens
- 256
- Speculative decoding
- None (standard decoding)
- Warm-ups
- 1
- Measured repetitions
- 3
- Timing method
- server durations aligned to client completion timestamps; aggregate-all uses client wall time
- Throughput spreads
- KV 8,192, 6 streams: PP mean 1,771.55 t/s (min 1,768.82, max 1,773.69, SD 2.03, n=3); TG mean 222.09 t/s (min 222.01, max 222.19, SD 0.08, n=3); All mean 619.81 t/s (min 619.37, max 620.29, SD 0.38, n=3)
- Measurement outcomes
- 8,192 tokens · 6 streams: measured
- Notes
- Automated llama-server KV-depth sweep (off, concurrency=6). Existing depth is contextTokens; PP=512, TG=256, warmups=1, repetitions=3, Q8_0 K/V. Arithmetic mean and population standard deviation of measured repetitions. PP/TG server durations are aligned to client completion timestamps for concurrent phase spans; aggregate-all includes SSH/HTTP wall time. Synthetic repeated-token prompts. Engine was reused as installed; no build was performed. Publication projection of reviewed source measurements; values and raw evidence unchanged. Single GPU — no splitting.
- Source article
- /benchmarks/
- Raw log path
research/benchmarks/logs/trpro/2026-10-14-approved-models-single-gpu/qwen36-iq4-xs-concurrency-isolated
Source 7 · 2026-10-15
- Source run ID
- pub-20261016-trpro-qwen36-iq4-xs-concurrency-2026-10-14-single-gpu-isolated-off-c7
- Page entry ID
- qwen3-6-35b-a3b-ud-iq4-xs-single-gpu-20261016-off-0f6923f607
- Entry sweep
- concurrency
- Raw entry label
- Qwen3.6 35B-A3B UD-IQ4_XS — Single GPU — no splitting
- Configuration label
- Single GPU — no splitting
- Measured date
- 2026-10-15
- Hardware
- 1× NVIDIA GeForce RTX 3090
- Runtime
- llamaCpp · version: 0.3.0-dev (build 50, commit fc6545d) built with GNU 15.2.0 for Linux x86_64
- Weight placement
- gpu-resident
- Observed devices
- CUDA0
- Requested devices
- CUDA0
- Split mode
- Single GPU — no splitting
- Placement detail
- Single GPU — no splitting
- KV-cache precision
- Q8_0
- Flash Attention
- Enabled
- Requested context per slot
- 9,024 tokens
- Actual context per slot
- 9,216 tokens
- Native model context
- 262,144 tokens
- Batch size
- 512
- Microbatch size
- 512
- Prompt tokens
- 512
- Generated tokens
- 256
- Speculative decoding
- None (standard decoding)
- Warm-ups
- 1
- Measured repetitions
- 3
- Timing method
- server durations aligned to client completion timestamps; aggregate-all uses client wall time
- Throughput spreads
- KV 8,192, 7 streams: PP mean 1,752.86 t/s (min 1,752.00, max 1,754.20, SD 0.96, n=3); TG mean 220.03 t/s (min 219.60, max 220.30, SD 0.31, n=3); All mean 620.37 t/s (min 619.22, max 621.10, SD 0.82, n=3)
- Measurement outcomes
- 8,192 tokens · 7 streams: measured
- Notes
- Automated llama-server KV-depth sweep (off, concurrency=7). Existing depth is contextTokens; PP=512, TG=256, warmups=1, repetitions=3, Q8_0 K/V. Arithmetic mean and population standard deviation of measured repetitions. PP/TG server durations are aligned to client completion timestamps for concurrent phase spans; aggregate-all includes SSH/HTTP wall time. Synthetic repeated-token prompts. Engine was reused as installed; no build was performed. Publication projection of reviewed source measurements; values and raw evidence unchanged. Single GPU — no splitting.
- Source article
- /benchmarks/
- Raw log path
research/benchmarks/logs/trpro/2026-10-14-approved-models-single-gpu/qwen36-iq4-xs-concurrency-isolated
Source 8 · 2026-10-15
- Source run ID
- pub-20261016-trpro-qwen36-iq4-xs-concurrency-2026-10-14-single-gpu-isolated-off-c8
- Page entry ID
- qwen3-6-35b-a3b-ud-iq4-xs-single-gpu-20261016-off-0f6923f607
- Entry sweep
- concurrency
- Raw entry label
- Qwen3.6 35B-A3B UD-IQ4_XS — Single GPU — no splitting
- Configuration label
- Single GPU — no splitting
- Measured date
- 2026-10-15
- Hardware
- 1× NVIDIA GeForce RTX 3090
- Runtime
- llamaCpp · version: 0.3.0-dev (build 50, commit fc6545d) built with GNU 15.2.0 for Linux x86_64
- Weight placement
- gpu-resident
- Observed devices
- CUDA0
- Requested devices
- CUDA0
- Split mode
- Single GPU — no splitting
- Placement detail
- Single GPU — no splitting
- KV-cache precision
- Q8_0
- Flash Attention
- Enabled
- Requested context per slot
- 9,024 tokens
- Actual context per slot
- 9,216 tokens
- Native model context
- 262,144 tokens
- Batch size
- 512
- Microbatch size
- 512
- Prompt tokens
- 512
- Generated tokens
- 256
- Speculative decoding
- None (standard decoding)
- Warm-ups
- 1
- Measured repetitions
- 3
- Timing method
- server durations aligned to client completion timestamps; aggregate-all uses client wall time
- Throughput spreads
- KV 8,192, 8 streams: PP mean 1,716.00 t/s (min 1,691.95, max 1,729.57, SD 17.05, n=3); TG mean 226.41 t/s (min 226.07, max 226.78, SD 0.29, n=3); All mean 641.51 t/s (min 639.78, max 643.14, SD 1.38, n=3)
- Measurement outcomes
- 8,192 tokens · 8 streams: measured
- Notes
- Automated llama-server KV-depth sweep (off, concurrency=8). Existing depth is contextTokens; PP=512, TG=256, warmups=1, repetitions=3, Q8_0 K/V. Arithmetic mean and population standard deviation of measured repetitions. PP/TG server durations are aligned to client completion timestamps for concurrent phase spans; aggregate-all includes SSH/HTTP wall time. Synthetic repeated-token prompts. Engine was reused as installed; no build was performed. Publication projection of reviewed source measurements; values and raw evidence unchanged. Single GPU — no splitting.
- Source article
- /benchmarks/
- Raw log path
research/benchmarks/logs/trpro/2026-10-14-approved-models-single-gpu/qwen36-iq4-xs-concurrency-isolated