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": 75323.49137226772,
-      "required_pages_per_second_upper_estimate": 0.005211053907712702
+      "required_token_positions_per_second": 75836.842243265,
+      "required_pages_per_second_upper_estimate": 0.005246568712106422
     },
     {
       "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": 9415.436421533464,
-      "required_pages_per_second_upper_estimate": 0.003942046613025553
+      "required_token_positions_per_second": 9479.605280408125,
+      "required_pages_per_second_upper_estimate": 0.003968912774238224
     },
     {
       "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": 18830.87284306693,
-      "required_pages_per_second_upper_estimate": 0.003921974209698829
+      "required_token_positions_per_second": 18959.21056081625,
+      "required_pages_per_second_upper_estimate": 0.003948703571812799
     },
     {
       "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": 28246.30926460039,
-      "required_pages_per_second_upper_estimate": 0.0016933231107958322
+      "required_token_positions_per_second": 28438.815841224376,
+      "required_pages_per_second_upper_estimate": 0.0017048635861239172
     },
     {
       "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": 28246.30926460039,
-      "required_pages_per_second_upper_estimate": 0.002196026302938565
+      "required_token_positions_per_second": 28438.815841224376,
+      "required_pages_per_second_upper_estimate": 0.0022109928425241366
     },
     {
       "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": 28246.30926460039,
-      "required_pages_per_second_upper_estimate": 0.3623133267223405
+      "required_token_positions_per_second": 28438.815841224376,
+      "required_pages_per_second_upper_estimate": 0.3647825943898151
     }
   ],
   "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": 0.10615653696066829,
-    "producer_p95_seconds": 0.29170451699974365,
-    "gpu_update_seconds": 0.9788181436733676,
-    "producer_batches_per_second": 9.420051073920268,
-    "gpu_batches_per_second": 1.0216402367115303,
-    "producer_headroom": 9.22051690548295,
+    "producer_batch_seconds": 0.0985632866326398,
+    "producer_p95_seconds": 0.2921264560027339,
+    "gpu_update_seconds": 0.9721923779935301,
+    "producer_batches_per_second": 10.145765570167628,
+    "gpu_batches_per_second": 1.0286030035165066,
+    "producer_headroom": 9.86363595622615,
     "sustainable": true,
-    "buffer_seconds": 60.68672490774879,
+    "buffer_seconds": 60.27592743559887,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 412416.9899146748,
-    "upload_bytes_per_second": 4987343.616074259,
+    "generation_bytes_per_second": 428380.58525201626,
+    "upload_bytes_per_second": 5078353.633928221,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 88.49508545952017,
+    "estimated_full_shards_per_hour": 91.92050140543333,
     "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": 1.439521827000135,
-      "bytes": 6360408,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.058846297000855,
       "bytes": 7513643,
@@ -562,6 +556,12 @@
       "bytes": 719591,
       "pages": 20,
       "shards": 1
+    },
+    {
+      "seconds": 1.395833352999034,
+      "bytes": 13910666,
+      "pages": 50,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -245,3 +245,4 @@
 {"step": 628900, "loss": 0.5531455874443054, "lr": 0.0001, "grad_norm": 0.17502718310119103, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92054220068, "tokens_per_second": 150524.7076351682, "data_wait_s": 1.763500040397048e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 629000, "loss": 0.5727645754814148, "lr": 0.0001, "grad_norm": 0.18562674109616814, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92068877363, "tokens_per_second": 141310.9472101476, "data_wait_s": 1.2853997759521008e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 629100, "loss": 0.5912374258041382, "lr": 0.0001, "grad_norm": 0.18060775264777926, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92083672772, "tokens_per_second": 153072.11397518427, "data_wait_s": 1.6065001545939595e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 629200, "loss": 0.5503384470939636, "lr": 0.0001, "grad_norm": 0.17518068853209828, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92098434220, "tokens_per_second": 152921.2282267415, "data_wait_s": 1.6170000890269876e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5bf5525904dbfe0f3fd3eebc14319459fe1024deb1aba3f4aff47031c008030f
-size 527226
+oid sha256:fc3a8aff32fba0b3a55a75bbf2077bd6a0aba0a37d2acaa3a4a61666e5b1374d
+size 529344