Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77221.3181014821,
-      "required_pages_per_second_upper_estimate": 0.005342349964404458
+      "required_token_positions_per_second": 76753.39920521603,
+      "required_pages_per_second_upper_estimate": 0.0053099782494393484
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9652.664762685263,
-      "required_pages_per_second_upper_estimate": 0.004041369165574722
+      "required_token_positions_per_second": 9594.174900652004,
+      "required_pages_per_second_upper_estimate": 0.004016880681748607
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19305.329525370526,
-      "required_pages_per_second_upper_estimate": 0.004020791024358543
+      "required_token_positions_per_second": 19188.349801304008,
+      "required_pages_per_second_upper_estimate": 0.003996427232798266
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28957.99428805579,
-      "required_pages_per_second_upper_estimate": 0.0017359875412718735
+      "required_token_positions_per_second": 28782.52470195601,
+      "required_pages_per_second_upper_estimate": 0.001725468407511736
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28957.99428805579,
-      "required_pages_per_second_upper_estimate": 0.002251356683140632
+      "required_token_positions_per_second": 28782.52470195601,
+      "required_pages_per_second_upper_estimate": 0.0022377146946305163
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28957.99428805579,
-      "required_pages_per_second_upper_estimate": 0.371442058055384
+      "required_token_positions_per_second": 28782.52470195601,
+      "required_pages_per_second_upper_estimate": 0.3691913226094587
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.06487789128129862,
-    "producer_p95_seconds": 0.23608135399990715,
-    "gpu_update_seconds": 0.9547622575298277,
-    "producer_batches_per_second": 15.413571252866154,
-    "gpu_batches_per_second": 1.0473811591455364,
-    "producer_headroom": 14.716296085983343,
+    "producer_batch_seconds": 0.4563886881015833,
+    "producer_p95_seconds": 0.26334320699970704,
+    "gpu_update_seconds": 0.9605828636054672,
+    "producer_batches_per_second": 2.19111478016611,
+    "gpu_batches_per_second": 1.0410346029353301,
+    "producer_headroom": 2.1047473100202256,
     "sustainable": true,
-    "buffer_seconds": 60.15002222437914,
+    "buffer_seconds": 60.51672040714443,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 366637.3428625618,
-    "upload_bytes_per_second": 3548393.8583439505,
+    "generation_bytes_per_second": 325023.2748029521,
+    "upload_bytes_per_second": 3545261.4651393583,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.67183889777795,
+    "estimated_full_shards_per_hour": 69.74242861811086,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
-    {
-      "seconds": 1.5859740610012523,
-      "bytes": 2563401,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.2778493570003775,
       "bytes": 10237665,
@@ -562,6 +556,12 @@
       "bytes": 13191953,
       "pages": 36,
       "shards": 4
+    },
+    {
+      "seconds": 1.2758062670000072,
+      "bytes": 1083353,
+      "pages": 30,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -185,3 +185,4 @@
 {"step": 861000, "loss": 0.5609935522079468, "lr": 3e-05, "grad_norm": 0.29852792585960003, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126025698838, "tokens_per_second": 151450.16261014933, "data_wait_s": 1.896000321721658e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 861100, "loss": 0.6365047693252563, "lr": 3e-05, "grad_norm": 0.3565185435318611, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126040404087, "tokens_per_second": 152822.69922314503, "data_wait_s": 1.1434000043664128e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 861200, "loss": 0.5374569892883301, "lr": 3e-05, "grad_norm": 0.3078046468393192, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126055039062, "tokens_per_second": 141664.39619205875, "data_wait_s": 1.2624001101357862e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 861300, "loss": 0.5280181169509888, "lr": 3e-05, "grad_norm": 0.2920627687392126, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126069751600, "tokens_per_second": 152517.52313176062, "data_wait_s": 1.7241000023204833e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:14e6e280e6997b390d3205f07238c7fa9c3f66dfe91afcadbd983987cc7725f2
-size 399363
+oid sha256:fa265bf562220834aefed0a2eed2bbb60cc0d99b58c887bee86e34a023b3bf41
+size 401481