Expose layer_roofline as loadable dataset config

README.md

@@ -32,6 +32,10 @@ configs:
     data_files:
       - split: summary
         path: mse_validation/summary.parquet
+  - config_name: layer_roofline
+    data_files:
+      - split: summary
+        path: layer_roofline/summary.parquet
 dataset_info:
   - config_name: trace_replay
     features:
@@ -328,15 +332,97 @@ dataset_info:
       - name: summary
         num_examples: 28
         num_bytes: 40940
+  - config_name: layer_roofline
+    features:
+      - name: record_type
+        dtype: string
+      - name: source_file
+        dtype: string
+      - name: model
+        dtype: string
+      - name: hardware
+        dtype: string
+      - name: engine
+        dtype: string
+      - name: dtype
+        dtype: string
+      - name: tensor_parallelism
+        dtype: int64
+      - name: phase
+        dtype: string
+      - name: batch_size
+        dtype: int64
+      - name: sequence_length
+        dtype: int64
+      - name: q_len
+        dtype: int64
+      - name: kv_len
+        dtype: int64
+      - name: component_name
+        dtype: string
+      - name: component_bound
+        dtype: string
+      - name: flops
+        dtype: float64
+      - name: bytes
+        dtype: float64
+      - name: operational_intensity_flop_per_byte
+        dtype: float64
+      - name: ridge_point_flop_per_byte
+        dtype: float64
+      - name: total_flops
+        dtype: float64
+      - name: total_bytes
+        dtype: float64
+      - name: overall_bound
+        dtype: string
+      - name: kernel_id
+        dtype: int64
+      - name: kernel_name
+        dtype: string
+      - name: block_size
+        dtype: string
+      - name: grid_size
+        dtype: string
+      - name: duration_us
+        dtype: float64
+      - name: compute_sm_throughput_pct
+        dtype: float64
+      - name: dram_throughput_pct
+        dtype: float64
+      - name: memory_throughput_pct
+        dtype: float64
+      - name: l1_tex_cache_throughput_pct
+        dtype: float64
+      - name: l2_cache_throughput_pct
+        dtype: float64
+      - name: sm_frequency_ghz
+        dtype: float64
+      - name: dram_frequency_ghz
+        dtype: float64
+      - name: artifact_path
+        dtype: string
+      - name: artifact_kind
+        dtype: string
+      - name: artifact_bytes
+        dtype: int64
+      - name: artifact_description
+        dtype: string
+      - name: notes
+        dtype: string
+    splits:
+      - name: summary
+        num_examples: 56
+        num_bytes: 27430
 ---
 
 # AgentPerfBench
 
-LLM inference benchmark: 3,392 main sweep rows measuring TTFT, TPOT, ITL, and throughput across 9 models, up to 14 GPU configurations, and 2 serving engines (vLLM 0.19.0, SGLang 0.5.9). All models served in BF16 except gpt-oss, which uses mxfp4 for projection weights. The dataset also includes 148,077 per-kernel NCU profiles and 28 curated MSE validation rows for the distributional synthetic replay generator.
+LLM inference benchmark: 3,392 main sweep rows measuring TTFT, TPOT, ITL, and throughput across 9 models, up to 14 GPU configurations, and 2 serving engines (vLLM 0.19.0, SGLang 0.5.9). All models served in BF16 except gpt-oss, which uses mxfp4 for projection weights. The dataset also includes 148,077 per-kernel NCU profiles, 28 curated MSE validation rows for the distributional synthetic replay generator, and 56 layer-roofline validation rows.
 
 ## Dataset configurations
 
-The dataset provides four configurations. `trace_replay` replays exact input/output sequences from recorded agent sessions. `synthetic_distributional` samples from statistical distributions fitted to those same workloads, trading fidelity for faster sweeps across the hardware matrix. `kernels_labeled` contains per-kernel Nsight Compute labels. `mse_validation` contains paired synthetic-vs-real validation runs and ablations for the final APC-aware synthetic generator.
+The dataset provides five configurations. `trace_replay` replays exact input/output sequences from recorded agent sessions. `synthetic_distributional` samples from statistical distributions fitted to those same workloads, trading fidelity for faster sweeps across the hardware matrix. `kernels_labeled` contains per-kernel Nsight Compute labels. `mse_validation` contains paired synthetic-vs-real validation runs and ablations for the final APC-aware synthetic generator. `layer_roofline` exposes the per-layer roofline validation artifacts as a loadable tabular subset.
 
 ### trace_replay (3,147 rows)
 
@@ -369,6 +455,12 @@ The headline SWE-bench C=5 source-locked cascade is:
 
 See `mse_validation/README.md`, `mse_validation/manifest.csv`, and `mse_validation/fidelity_deltas.csv` for the filtering rule and raw JSON pointers.
 
+### layer_roofline (56 rows)
+
+Tabular view over the per-layer roofline evidence for Llama-3.1-8B on H100. This subset combines analytical component OI rows, selected NCU kernel summary rows, and an artifact manifest pointing to the raw `per_layer_oi_cf/` evidence files.
+
+Record types: `analytical_total`, `analytical_component`, `ncu_kernel`, `artifact`.
+
 ### Concurrency filtering
 
 The benchmark harness capped actual concurrent connections at the session pool size. Rows where declared concurrency exceeded the pool were excluded:
@@ -381,6 +473,7 @@ The benchmark harness capped actual concurrent connections at the session pool s
 | trace_replay | 3,147 |
 | synthetic_distributional | 245 |
 | mse_validation | 28 |
+| layer_roofline | 56 |
 | kernels_labeled | 148,077 |
 
 ### Failed requests
@@ -458,7 +551,7 @@ Each row in the serving `summary.parquet` configs:
 from datasets import load_dataset
 
 ds = load_dataset("agent-perf-bench/AgentPerfBench", "trace_replay")
-# or "synthetic_distributional", "kernels_labeled", "mse_validation"
+# or "synthetic_distributional", "kernels_labeled", "mse_validation", "layer_roofline"
 ```
 
 ## Benchmark methodology
@@ -474,7 +567,7 @@ ds = load_dataset("agent-perf-bench/AgentPerfBench", "trace_replay")
 ## Future releases
 
 - Full per-request and multi-turn granularity data for the main sweep (pending raw JSON availability from collection infrastructure). Curated raw JSONs are included for `mse_validation`.
-- Additional per-kernel roofline profiles beyond the included `kernels_labeled` and `per_layer_oi_cf` artifacts.
+- Additional per-kernel roofline profiles beyond the included `kernels_labeled`, `layer_roofline`, and raw `per_layer_oi_cf` artifacts.
 - This is version 1.0. Updates will be tagged with semantic versions.
 
 ## Intended uses

layer_roofline/summary.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1f57b62366d8dd0c2eda436f1d559257f409acf47bbf52531b538de767cebd19
+size 27430