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": 75641.5982275477,
-      "required_pages_per_second_upper_estimate": 0.005233061278070567
+      "required_token_positions_per_second": 76139.62741258068,
+      "required_pages_per_second_upper_estimate": 0.005267516092678067
     },
     {
       "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": 9455.199778443462,
-      "required_pages_per_second_upper_estimate": 0.003958694700210454
+      "required_token_positions_per_second": 9517.453426572585,
+      "required_pages_per_second_upper_estimate": 0.0039847590026781165
     },
     {
       "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": 18910.399556886925,
-      "required_pages_per_second_upper_estimate": 0.003938537526927057
+      "required_token_positions_per_second": 19034.90685314517,
+      "required_pages_per_second_upper_estimate": 0.003964469113259443
     },
     {
       "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": 28365.599335330386,
-      "required_pages_per_second_upper_estimate": 0.001700474368390704
+      "required_token_positions_per_second": 28552.360279717752,
+      "required_pages_per_second_upper_estimate": 0.0017116704018392763
     },
     {
       "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": 28365.599335330386,
-      "required_pages_per_second_upper_estimate": 0.0022053005812362535
+      "required_token_positions_per_second": 28552.360279717752,
+      "required_pages_per_second_upper_estimate": 0.002219820423187793
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28365.599335330386,
-      "required_pages_per_second_upper_estimate": 0.36384345166596616
+      "required_token_positions_per_second": 28552.360279717752,
+      "required_pages_per_second_upper_estimate": 0.36623902053228863
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.1333228127109578,
-    "producer_p95_seconds": 0.2763026920001721,
-    "gpu_update_seconds": 0.9747017742566578,
-    "producer_batches_per_second": 7.500591831706908,
-    "gpu_batches_per_second": 1.0259548370706881,
-    "producer_headroom": 7.310840166339718,
+    "producer_batch_seconds": 0.10305958707809282,
+    "producer_p95_seconds": 0.23571275699941907,
+    "gpu_update_seconds": 0.968326251460193,
+    "producer_batches_per_second": 9.703124457914388,
+    "gpu_batches_per_second": 1.0327097902096989,
+    "producer_headroom": 9.395790133783956,
     "sustainable": true,
-    "buffer_seconds": 60.431510003912784,
+    "buffer_seconds": 60.036227590531965,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 245601.3766431884,
-    "upload_bytes_per_second": 4970547.649847886,
+    "generation_bytes_per_second": 264425.0754298608,
+    "upload_bytes_per_second": 5004917.496550926,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 52.700338120191,
+    "estimated_full_shards_per_hour": 56.739465686529805,
     "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": 1.0984560849992704,
-      "bytes": 1168159,
-      "pages": 35,
-      "shards": 1
-    },
     {
       "seconds": 1.9164282879992243,
       "bytes": 13374744,
@@ -562,6 +556,12 @@
       "bytes": 5700613,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 1.7661054199998034,
+      "bytes": 7454533,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -172,3 +172,4 @@
 {"step": 737200, "loss": 0.5161051154136658, "lr": 5e-05, "grad_norm": 0.2209857690773546, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107905571719, "tokens_per_second": 148907.93041877376, "data_wait_s": 1.9220999092794955e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 737300, "loss": 0.5317947268486023, "lr": 5e-05, "grad_norm": 0.21974134659607758, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107920257550, "tokens_per_second": 150195.63340250257, "data_wait_s": 2.112099900841713e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 737400, "loss": 0.5872620344161987, "lr": 5e-05, "grad_norm": 0.22753151759390347, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107935006358, "tokens_per_second": 151643.86668693027, "data_wait_s": 1.6305002645822242e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 737500, "loss": 0.49660512804985046, "lr": 5e-05, "grad_norm": 0.22140448534167811, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107949720133, "tokens_per_second": 150500.157676661, "data_wait_s": 2.0002000383101404e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c7b7aee8ad9ba4db7e8a3d793954a054bb62bf5ebd2c2bd6038b737b9593bf49
-size 371046
+oid sha256:1d21579a2333da62b3dbee415eaa3a259b26857f8f00efd1a2cf3ab501126042
+size 373164