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": 76561.19447506017,
-      "required_pages_per_second_upper_estimate": 0.005296681080230748
+      "required_token_positions_per_second": 77014.92489138205,
+      "required_pages_per_second_upper_estimate": 0.00532807120323151
     },
     {
       "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": 9570.149309382521,
-      "required_pages_per_second_upper_estimate": 0.004006821668395472
+      "required_token_positions_per_second": 9626.865611422756,
+      "required_pages_per_second_upper_estimate": 0.004030567599688652
     },
     {
       "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": 19140.298618765042,
-      "required_pages_per_second_upper_estimate": 0.003986419438670299
+      "required_token_positions_per_second": 19253.731222845512,
+      "required_pages_per_second_upper_estimate": 0.004010044458681326
     },
     {
       "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": 28710.447928147565,
-      "required_pages_per_second_upper_estimate": 0.001721147514976781
+      "required_token_positions_per_second": 28880.59683426827,
+      "required_pages_per_second_upper_estimate": 0.0017313476820963807
     },
     {
       "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": 28710.447928147565,
-      "required_pages_per_second_upper_estimate": 0.002232111042499133
+      "required_token_positions_per_second": 28880.59683426827,
+      "required_pages_per_second_upper_estimate": 0.0022453393715440736
     },
     {
       "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": 28710.447928147565,
-      "required_pages_per_second_upper_estimate": 0.3682667991450541
+      "required_token_positions_per_second": 28880.59683426827,
+      "required_pages_per_second_upper_estimate": 0.37044928662110893
     }
   ],
   "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": 0.31144588858603584,
-    "producer_p95_seconds": 0.3554697510044207,
-    "gpu_update_seconds": 0.9629943799272478,
-    "producer_batches_per_second": 3.210830634303761,
-    "gpu_batches_per_second": 1.0384276594382076,
-    "producer_headroom": 3.0920118557327623,
+    "producer_batch_seconds": 0.1260480845623988,
+    "producer_p95_seconds": 0.36400515000423184,
+    "gpu_update_seconds": 0.9573209362208979,
+    "producer_batches_per_second": 7.933480333887663,
+    "gpu_batches_per_second": 1.0445817720727817,
+    "producer_headroom": 7.594886820727419,
     "sustainable": true,
-    "buffer_seconds": 60.66864593541661,
+    "buffer_seconds": 60.31121898191657,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 292173.39620120096,
-    "upload_bytes_per_second": 4467479.035478239,
+    "generation_bytes_per_second": 291431.1194744689,
+    "upload_bytes_per_second": 4451672.356310293,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.69360937621137,
+    "estimated_full_shards_per_hour": 62.53433406997252,
     "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": 1.2590785899956245,
-      "bytes": 6075199,
-      "pages": 38,
-      "shards": 2
-    },
     {
       "seconds": 1.3119273870033794,
       "bytes": 2422074,
@@ -562,6 +556,12 @@
       "bytes": 11465498,
       "pages": 39,
       "shards": 4
+    },
+    {
+      "seconds": 1.7269667380023748,
+      "bytes": 6633188,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -424,3 +424,4 @@
 {"step": 646800, "loss": 0.5721449255943298, "lr": 0.0001, "grad_norm": 0.17860234885716397, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94675145228, "tokens_per_second": 151517.47726566327, "data_wait_s": 1.3228003808762878e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 646900, "loss": 0.557778000831604, "lr": 0.0001, "grad_norm": 0.1944920305595768, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94689630493, "tokens_per_second": 149477.76799090422, "data_wait_s": 1.7053003830369562e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 647000, "loss": 0.5787025690078735, "lr": 0.0001, "grad_norm": 0.1755085000784506, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94704061158, "tokens_per_second": 149091.65776771074, "data_wait_s": 9.91400156635791e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 647100, "loss": 0.5414162874221802, "lr": 0.0001, "grad_norm": 0.18974881924892584, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94718643414, "tokens_per_second": 151372.93778320594, "data_wait_s": 1.3470002159010619e-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 sha256:0bfe234d63bc59102629ee043eb688a0bf3e8bce6f93c147c6beaf561b53f42e
-size 909480
+oid sha256:61ca4468662ef0c4d30a8d740b0f367f4bf511a23555322a15d6a91723bfc1f5
+size 911598