Results from neuronxcc-2.28.7631.0+8f7b0cf5

README.md

@@ -32,7 +32,9 @@ from the prompt `"The future of AI is"`, sent as a user message for chat
 models. The model is loaded with `tp_plan="auto"`, or expert-parallel for
 mixture-of-experts models without a tensor-parallel plan, at the smallest
 power-of-two degree that fits its bf16 weights. Generation uses
-`torch.compile(backend="neuron")` with a static cache and chunked prefill.
+`torch.compile(backend="neuron")` with a static cache and chunked prefill. An
+encoder-decoder model runs its encoder once on the prompt, then generates from
+its decoder, at TP=1.
 
 For now, only models that fit on 4 Trainium2 devices are tested: the degree is
 at most 16.
@@ -44,6 +46,7 @@ at most 16.
 | `FAIL(gen)` | `generate()` raised. |
 | `FAIL(output)` | `generate()` returned, but the output check failed. |
 | `FAIL(other)` | The run crashed, or the model's extra Python packages could not be installed. |
+| `EXCLUDED` | Not tested: the sweep's selection left the model out, for the reason in `notes`. A later run of the model replaces this row. |
 
 The output check compares a TP=1 run with a CPU run of the same checkpoint
 and prompt: the tokens must match, or differ first at a near-tie. At higher
@@ -52,9 +55,12 @@ the text is fluent English. `PASS` says nothing about throughput or about
 longer generations.
 
 The models are the top `text-generation` and `image-text-to-text` models on
-the Hub by downloads and trending, reselected before each sweep. Quantized
-checkpoints and models that need `trust_remote_code` are left out. A model
-that leaves the list keeps the results it already has.
+the Hub by downloads and trending, reselected before each sweep: the 100
+best-ranked form the selection. Each sweep also tests the next 100
+best-ranked of these tasks and of `translation` and `summarization`,
+encoder-decoder models included. Quantized checkpoints and models that need
+`trust_remote_code` are left out. A model that leaves the list keeps the
+results it already has.
 
 ## Layout
 
@@ -66,17 +72,18 @@ data/neuronxcc-<version>/<library>-<version>.parquet
 
 Each sweep tests the current Neuron SDK with the library version of the day,
 so most pairs are never tested. A model missing from a pair was not tested
-there, which does not mean it fails.
+there, which does not mean it fails. A model the sweep left out has an
+`EXCLUDED` row instead, with no `tp` or `runner`.
 
-`selection/<library>.csv` lists the models the latest sweep of a library
-selected, best-ranked first, with the downloads and the Hub ranks they were
-selected by. `excluded/<library>.csv` lists the models that selection left out,
-with the reason and the value of the rule that matched.
+`selection/<library>.csv` lists the 100 best-ranked models of the latest sweep
+of a library, best-ranked first, with the downloads and the Hub ranks they were
+selected by.
 
 | Column | Content |
 |---|---|
-| `model_id`, `library`, `tp`, `runner` | The model, the library that loads it, its tensor- or expert-parallel degree, and the load class (`run_dense`: `AutoModelForCausalLM`, `run_multimodal`: `AutoModelForImageTextToText`, `run_dense_ep`: expert-parallel). |
-| `result`, `notes` | The outcome, and the start of the generated text or of the error. |
+| `model_id`, `library`, `tp`, `runner` | The model, the library that loads it, its tensor- or expert-parallel degree, and the load class (`run_dense`: `AutoModelForCausalLM`, `run_multimodal`: `AutoModelForImageTextToText`, `run_seq2seq`: `AutoModelForSeq2SeqLM`, `run_dense_ep`: expert-parallel). |
+| `result`, `notes` | The outcome, and the start of the generated text or of the error, or why the model was excluded. |
+| `exclusion` | The category of the rule that excluded an `EXCLUDED` model, such as `quantized` or `trust-remote-code`. |
 | `error` | The full exception message of a `FAIL(load)` or `FAIL(gen)` run. |
 | `cpu_match` | TP=1 only: `Y` if the tokens match a CPU run, `~ near-tie …` if they differ only at a near-tie, `N …` if not. |
 | `english` | The judge's verdict, `Y` or `N`, when the CPU comparison did not run. |

data/neuronxcc-2.28.7631.0+8f7b0cf5/transformers-5.20.2.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:eabc5e0c157025a8a975e5a18043aff6fc3f5f1d9aa1c50880532a48a8466313
-size 177518
+oid sha256:4ad9fa7dde19b11ed0a995e1d728afff096fecb7442c4c76615c6331d39787bf
+size 358630

selection/transformers.csv

@@ -1,101 +1,101 @@
 runner,tp,model_id,downloads,rank_downloads,rank_trending
-run_dense,1,Qwen/Qwen3-0.6B,28228809,1,
-run_dense,4,Qwen/Qwen3.8-27B,7729432,4,1
-run_dense_ep,8,google/gemma-4-26B-A4B-it,9789729,1,7
-run_dense_ep,8,LiquidAI/LFM2.5-24B-A2B,16939,,2
-run_dense,1,Qwen/Qwen3-4B,11600540,2,
-run_dense,4,google/gemma-4-31B-it,8827508,2,5
-run_dense,4,Qwen/Qwen3-14B,9034965,3,
-run_dense,1,Qwen/Qwen3.5-4B,7792311,3,
-run_dense,2,Qwen/Qwen3-8B,8584494,4,
-run_dense,2,Qwen/Qwen2.5-7B-Instruct,6892012,5,
-run_dense,2,Qwen/Qwen3.5-9B,7048019,5,6
-run_dense,2,meta-llama/Meta-Llama-3-8B-Instruct,429637,53,5
-run_dense,1,Qwen/Qwen2.5-0.5B-Instruct,5663553,6,
-run_multimodal,2,Qwen/Qwen2.5-VL-7B-Instruct,5124182,6,
-run_dense,2,meta-llama/Llama-3.1-8B-Instruct,5255644,7,6
-run_multimodal,2,Qwen/Qwen3-VL-8B-Instruct,5026100,7,
-run_dense,1,Qwen/Qwen2.5-1.5B-Instruct,5096561,8,
-run_dense,2,google/gemma-4-E4B-it,4076573,8,9
-run_dense,1,openbmb/MiniCPM5-2B,1322537,19,8
-run_dense_ep,8,Qwen/Qwen3.6-35B-A3B,3682645,9,
-run_dense,1,meta-llama/Llama-3.2-1B-Instruct,4633884,9,
-run_dense,1,meta-llama/Llama-3.2-3B-Instruct,1110545,24,9
-run_dense,1,Qwen/Qwen3-1.7B,4238898,10,
-run_multimodal,1,Qwen/Qwen3-VL-4B-Instruct,3602828,10,
-run_dense,2,meta-llama/Llama-3.1-8B,642867,37,10
-run_dense,1,Qwen/Qwen2.5-3B-Instruct,4071966,11,
-run_dense,1,Qwen/Qwen3.5-0.8B,3297603,11,
-run_dense,8,Qwen/Qwen3-32B,2837077,12,
-run_dense,1,Qwen/Qwen3.5-2B,2861330,12,
-run_dense,1,Qwen/Qwen3-4B-Instruct-2507,2764795,13,
-run_dense,1,google/gemma-4-E2B-it,2853063,13,
-run_dense,1,HuggingFaceTB/SmolLM2-135M,1746447,14,
-run_dense,4,Qwen/Qwen3.6-27B,2353721,15,
-run_dense,2,mistralai/Mistral-7B-Instruct-v0.2,1745584,15,
-run_dense,4,Qwen/Qwen2.5-Coder-14B-Instruct,1687825,16,
-run_multimodal,1,Qwen/Qwen3-VL-2B-Instruct,2037547,16,
-run_dense,2,Qwen/Qwen2.5-Coder-7B-Instruct,1548988,17,
-run_multimodal,1,Qwen/Qwen2.5-VL-3B-Instruct,1804143,17,
-run_dense,4,Qwen/Qwen3.5-27B,1720102,18,
-run_dense,1,google/gemma-3-1b-it,1396612,18,
-run_multimodal,1,HuggingFaceTB/SmolVLM2-500M-Video-Instruct,1443498,20,
-run_dense,1,deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B,1319824,20,
-run_dense,8,Qwen/Qwen2.5-32B-Instruct,1304459,21,
-run_dense_ep,8,Qwen/Qwen3.5-35B-A3B,1416929,21,
-run_multimodal,1,Qwen/Qwen2-VL-2B-Instruct,1183922,22,
-run_dense,1,Qwen/Qwen2.5-0.5B,1270098,22,
-run_dense,1,HuggingFaceTB/SmolLM2-135M-Instruct,1250316,23,
-run_dense,1,google/gemma-3-4b-it,1059693,23,
-run_multimodal,2,llava-hf/llava-1.5-7b-hf,1019993,24,
-run_multimodal,8,Qwen/Qwen2.5-VL-32B-Instruct,818811,25,
-run_dense,4,Qwen/Qwen3-30B-A3B,1098061,25,
-run_dense,4,Qwen/Qwen3-30B-A3B-Instruct-2507,1090144,26,
-run_dense,1,Qwen/Qwen3.5-4B-Base,759153,26,
-run_dense,4,zai-org/GLM-4.7-Flash,1053541,27,
-run_multimodal,2,Qwen/Qwen2-VL-7B-Instruct,662169,28,
-run_dense,4,Qwen/Qwen2.5-14B-Instruct,862411,28,
-run_dense,2,google/gemma-4-E4B,530802,29,
-run_dense,1,HuggingFaceTB/SmolLM3-3B,743150,30,
-run_dense,1,Qwen/Qwen3.5-0.8B-Base,512001,30,
-run_dense,2,Qwen/Qwen2.5-7B,740387,31,
-run_dense,1,google/medgemma-4b-it,500957,31,
-run_dense,2,Qwen/Qwen3-8B-Base,739797,32,
-run_dense,2,ornith-ai/Ornith-1.5-9B,455440,32,
-run_dense,2,Qwen/Qwen3.5-9B-Base,371264,33,
-run_dense,2,google/gemma-2-9b-it,682640,34,
-run_dense,8,google/gemma-3-27b-it,370529,34,
-run_dense,1,Qwen/Qwen2.5-1.5B,663517,35,
-run_dense,2,google/gemma-3-12b-it,352422,35,
-run_dense,1,Qwen/Qwen3-0.6B-Base,645510,36,
-run_dense,1,Qwen/Qwen3.5-2B-Base,323397,36,
-run_multimodal,1,ibm-granite/granite-docling-258M,303152,37,
-run_dense,8,Qwen/Qwen2.5-Coder-32B-Instruct,642833,38,
-run_multimodal,2,llava-hf/llava-v1.6-mistral-7b-hf,291265,38,
-run_multimodal,2,datalab-to/chandra,287425,39,
-run_dense,1,ibm-granite/granite-4.1-3b,630456,39,
-run_dense,1,Qwen/Qwen3-4B-Thinking-2507,614385,40,
-run_dense,4,google/gemma-4-31B,285427,40,
-run_dense,1,Qwen/Qwen3-4B-Base,599868,41,
-run_multimodal,1,llava-hf/llava-onevision-qwen2-0.5b-ov-hf,280310,41,
-run_dense,4,Qwen/Qwen3-Coder-30B-A3B-Instruct,576798,42,
-run_multimodal,4,Qwen/Qwen3-VL-30B-A3B-Instruct,272614,42,
-run_multimodal,1,google/paligemma-3b-pt-224,267262,43,
-run_dense,1,meta-llama/Llama-3.2-1B,573597,43,
-run_dense,1,Qwen/Qwen3-1.7B-Base,559875,44,
-run_dense_ep,8,ornith-ai/Ornith-1.5-35B-A3B,260853,44,
-run_multimodal,1,HuggingFaceTB/SmolVLM-256M-Instruct,244267,45,
-run_dense,2,meta-llama/Llama-2-7b-hf,536187,45,
-run_dense_ep,16,Qwen/Qwen3-Coder-Next,521313,46,
-run_multimodal,8,Qwen/Qwen3-VL-32B-Instruct,210971,47,
-run_dense,2,allenai/Olmo-3-7B-Think,490092,47,
-run_dense,1,Qwen/Qwen2.5-Coder-3B-Instruct,470485,48,
-run_multimodal,2,mistral-experimental/pixtral-12b,199590,48,
-run_dense,1,google/gemma-4-E2B,197286,49,
-run_dense,1,microsoft/phi-2,470309,49,
-run_dense,2,deepseek-ai/deepseek-coder-7b-instruct-v1.5,458958,50,
-run_multimodal,1,openbmb/MiniCPM-V-4.6,186279,50,
-run_dense,1,google/gemma-3-4b-pt,176729,51,
-run_multimodal,2,ByteDance-Seed/UI-TARS-1.5-7B,174751,52,
-run_dense,2,swiss-ai/Apertus-8B-Instruct-2509,429909,52,
-run_dense,1,Qwen/Qwen2.5-Coder-1.5B-Instruct,415940,54,
+run_dense,1,Qwen/Qwen3-0.6B,26678187,1,
+run_dense,4,Qwen/Qwen3.8-27B,7758670,3,1
+run_dense_ep,8,google/gemma-4-26B-A4B-it,9441586,1,8
+run_dense,1,Qwen/Qwen3-4B,11578451,2,
+run_dense,4,google/gemma-4-31B-it,8630477,2,5
+run_dense_ep,8,LiquidAI/LFM2.5-24B-A2B,17637,,3
+run_dense,4,Qwen/Qwen3-14B,9211389,3,
+run_dense,2,Qwen/Qwen3-8B,8298786,4,
+run_dense,1,Qwen/Qwen3.5-4B,7630082,4,
+run_dense,2,Qwen/Qwen2.5-7B-Instruct,6579331,5,
+run_dense,2,Qwen/Qwen3.5-9B,6689005,5,6
+run_dense,2,meta-llama/Meta-Llama-3-8B-Instruct,396532,54,5
+run_dense,1,Qwen/Qwen2.5-0.5B-Instruct,5488241,6,
+run_multimodal,2,Qwen/Qwen2.5-VL-7B-Instruct,5033876,6,
+run_dense,2,meta-llama/Llama-3.1-8B-Instruct,5046973,7,6
+run_multimodal,2,Qwen/Qwen3-VL-8B-Instruct,4785534,7,
+run_dense,1,meta-llama/Llama-3.2-3B-Instruct,1089842,24,7
+run_dense,1,Qwen/Qwen2.5-1.5B-Instruct,4938875,8,
+run_dense,2,google/gemma-4-E4B-it,3994620,8,9
+run_dense,1,openbmb/MiniCPM5-2B,1290255,19,8
+run_dense_ep,8,Qwen/Qwen3.6-35B-A3B,3717569,9,10
+run_dense,1,meta-llama/Llama-3.2-1B-Instruct,4303294,9,
+run_dense,1,Qwen/Qwen3-1.7B,4185973,10,
+run_multimodal,1,Qwen/Qwen3-VL-4B-Instruct,3600122,10,
+run_dense,2,meta-llama/Llama-3.1-8B,630280,36,10
+run_dense,1,Qwen/Qwen2.5-3B-Instruct,3940831,11,
+run_dense,1,Qwen/Qwen3.5-0.8B,3317268,11,
+run_dense,8,Qwen/Qwen3-32B,2751050,12,
+run_dense,1,Qwen/Qwen3.5-2B,2898413,12,
+run_dense,1,Qwen/Qwen3-4B-Instruct-2507,2670062,13,
+run_dense,1,google/gemma-4-E2B-it,2848427,13,
+run_dense,1,HuggingFaceTB/SmolLM2-135M,1808635,14,
+run_dense,4,Qwen/Qwen3.6-27B,2350861,15,
+run_dense,2,mistralai/Mistral-7B-Instruct-v0.2,1745610,15,
+run_dense,4,Qwen/Qwen2.5-Coder-14B-Instruct,1674011,16,
+run_multimodal,1,Qwen/Qwen3-VL-2B-Instruct,1960387,16,
+run_dense,2,Qwen/Qwen2.5-Coder-7B-Instruct,1532215,17,
+run_multimodal,1,Qwen/Qwen2.5-VL-3B-Instruct,1743974,17,
+run_dense,1,deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B,1309873,18,
+run_dense,4,Qwen/Qwen3.5-27B,1689285,19,
+run_multimodal,1,HuggingFaceTB/SmolVLM2-500M-Video-Instruct,1456185,20,
+run_dense,8,Qwen/Qwen2.5-32B-Instruct,1280269,20,
+run_dense,1,Qwen/Qwen2.5-0.5B,1233535,21,
+run_dense_ep,8,Qwen/Qwen3.5-35B-A3B,1410542,21,
+run_multimodal,1,Qwen/Qwen2-VL-2B-Instruct,1161836,22,
+run_dense,1,google/gemma-3-1b-it,1217096,22,
+run_dense,1,HuggingFaceTB/SmolLM2-135M-Instruct,1192475,23,
+run_dense,1,google/gemma-3-4b-it,1025839,23,
+run_multimodal,2,llava-hf/llava-1.5-7b-hf,976357,24,
+run_multimodal,8,Qwen/Qwen2.5-VL-32B-Instruct,812677,25,
+run_dense,4,Qwen/Qwen3-30B-A3B-Instruct-2507,1083290,25,
+run_dense,4,Qwen/Qwen3-30B-A3B,1046589,26,
+run_dense,1,Qwen/Qwen3.5-4B-Base,748130,26,
+run_dense,4,zai-org/GLM-4.7-Flash,1009048,27,
+run_multimodal,2,Qwen/Qwen2-VL-7B-Instruct,650870,28,
+run_dense,4,Qwen/Qwen2.5-14B-Instruct,826546,28,
+run_dense,2,google/gemma-4-E4B,516887,29,
+run_dense,1,HuggingFaceTB/SmolLM3-3B,731019,30,
+run_dense,1,Qwen/Qwen3.5-0.8B-Base,495395,30,
+run_dense,2,Qwen/Qwen3-8B-Base,725705,31,
+run_dense,1,google/medgemma-4b-it,471909,31,
+run_dense,2,ornith-ai/Ornith-1.5-9B,460907,32,
+run_dense,2,Qwen/Qwen2.5-7B,700938,33,
+run_dense,2,Qwen/Qwen3.5-9B-Base,366124,33,
+run_dense,1,Qwen/Qwen2.5-1.5B,654537,34,
+run_dense,8,google/gemma-3-27b-it,355849,34,
+run_dense,1,Qwen/Qwen3-0.6B-Base,648929,35,
+run_dense,2,google/gemma-3-12b-it,345595,35,
+run_dense,1,Qwen/Qwen3.5-2B-Base,321403,36,
+run_dense,1,ibm-granite/granite-4.1-3b,620154,37,
+run_multimodal,1,ibm-granite/granite-docling-258M,304046,37,
+run_dense,1,Qwen/Qwen3-4B-Thinking-2507,607738,38,
+run_multimodal,2,llava-hf/llava-v1.6-mistral-7b-hf,281587,38,
+run_dense,4,Qwen/Qwen3-Coder-30B-A3B-Instruct,602923,39,
+run_multimodal,4,Qwen/Qwen3-VL-30B-A3B-Instruct,277219,39,
+run_dense,8,Qwen/Qwen2.5-Coder-32B-Instruct,588474,40,
+run_multimodal,2,datalab-to/chandra,276650,40,
+run_dense,1,Qwen/Qwen3-4B-Base,587349,41,
+run_multimodal,1,llava-hf/llava-onevision-qwen2-0.5b-ov-hf,264246,41,
+run_dense,2,google/gemma-2-9b-it,575723,42,
+run_dense_ep,8,ornith-ai/Ornith-1.5-35B-A3B,263794,42,
+run_dense,4,google/gemma-4-31B,263140,43,
+run_dense,1,meta-llama/Llama-3.2-1B,563629,43,
+run_dense,1,Qwen/Qwen3-1.7B-Base,555358,44,
+run_multimodal,1,google/paligemma-3b-pt-224,260462,44,
+run_multimodal,1,HuggingFaceTB/SmolVLM-256M-Instruct,231254,45,
+run_dense_ep,16,Qwen/Qwen3-Coder-Next,516351,45,
+run_dense,2,meta-llama/Llama-2-7b-hf,499370,46,
+run_dense,2,allenai/Olmo-3-7B-Think,489484,47,
+run_dense,1,google/gemma-3-4b-pt,206334,47,
+run_multimodal,8,Qwen/Qwen3-VL-32B-Instruct,205367,48,
+run_dense,1,Qwen/Qwen2.5-Coder-3B-Instruct,455605,49,
+run_multimodal,2,mistral-experimental/pixtral-12b,199747,49,
+run_dense,1,google/gemma-4-E2B,191552,50,
+run_dense,1,microsoft/phi-2,451167,50,
+run_dense,2,deepseek-ai/deepseek-coder-7b-instruct-v1.5,440158,51,
+run_multimodal,1,openbmb/MiniCPM-V-4.6,181313,51,
+run_multimodal,2,ByteDance-Seed/UI-TARS-1.5-7B,157799,52,
+run_dense,2,swiss-ai/Apertus-8B-Instruct-2509,422916,52,
+run_dense,1,Qwen/Qwen2.5-Coder-1.5B-Instruct,408481,53,