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": 77289.96157501306,
-      "required_pages_per_second_upper_estimate": 0.005347098879178132
+      "required_token_positions_per_second": 77205.38072069423,
+      "required_pages_per_second_upper_estimate": 0.005341247379421735
     },
     {
       "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": 9661.245196876633,
-      "required_pages_per_second_upper_estimate": 0.004044961614190807
+      "required_token_positions_per_second": 9650.672590086779,
+      "required_pages_per_second_upper_estimate": 0.004040535084509031
     },
     {
       "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": 19322.490393753265,
-      "required_pages_per_second_upper_estimate": 0.004024365180680135
+      "required_token_positions_per_second": 19301.345180173557,
+      "required_pages_per_second_upper_estimate": 0.004019961190328263
     },
     {
       "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": 28983.735590629898,
-      "required_pages_per_second_upper_estimate": 0.001737530693056498
+      "required_token_positions_per_second": 28952.017770260336,
+      "required_pages_per_second_upper_estimate": 0.0017356292581556503
     },
     {
       "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": 28983.735590629898,
-      "required_pages_per_second_upper_estimate": 0.0022533579561917452
+      "required_token_positions_per_second": 28952.017770260336,
+      "required_pages_per_second_upper_estimate": 0.0022508920351699594
     },
     {
       "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": 28983.735590629898,
-      "required_pages_per_second_upper_estimate": 0.3717722398459473
+      "required_token_positions_per_second": 28952.017770260336,
+      "required_pages_per_second_upper_estimate": 0.37136539770218874
     }
   ],
   "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": 0.06542800232823254,
-    "producer_p95_seconds": 0.23194343200157164,
-    "gpu_update_seconds": 0.9539143052677542,
-    "producer_batches_per_second": 15.283975735393874,
-    "gpu_batches_per_second": 1.0483121958416484,
-    "producer_headroom": 14.579603095357461,
+    "producer_batch_seconds": 0.11433480384374661,
+    "producer_p95_seconds": 0.22744864800006326,
+    "gpu_update_seconds": 0.9549593475450326,
+    "producer_batches_per_second": 8.746243194387512,
+    "gpu_batches_per_second": 1.0471649945840689,
+    "producer_headroom": 8.35230669438248,
     "sustainable": true,
-    "buffer_seconds": 60.09660123186852,
+    "buffer_seconds": 60.162438895337054,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 269009.81653519406,
-    "upload_bytes_per_second": 5594367.3685591575,
+    "generation_bytes_per_second": 290622.43947995064,
+    "upload_bytes_per_second": 5552795.949776833,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.72324440042368,
+    "estimated_full_shards_per_hour": 62.3608101682557,
     "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": 1.0036822970005232,
-      "bytes": 2451346,
-      "pages": 32,
-      "shards": 2
-    },
     {
       "seconds": 1.0700342689997342,
       "bytes": 10599218,
@@ -562,6 +556,12 @@
       "bytes": 13564102,
       "pages": 48,
       "shards": 4
+    },
+    {
+      "seconds": 1.359001171998898,
+      "bytes": 1213849,
+      "pages": 32,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -119,3 +119,4 @@
 {"step": 616300, "loss": 0.557181179523468, "lr": 0.0001, "grad_norm": 0.1463737183512701, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90208205468, "tokens_per_second": 140816.81178792083, "data_wait_s": 1.473700103815645e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 616400, "loss": 0.5112867951393127, "lr": 0.0001, "grad_norm": 0.14831743242015838, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90222981911, "tokens_per_second": 152916.38834635567, "data_wait_s": 1.738600076350849e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 616500, "loss": 0.5556550025939941, "lr": 0.0001, "grad_norm": 0.15794927870746733, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90237762388, "tokens_per_second": 153809.371311569, "data_wait_s": 9.044999387697317e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 616600, "loss": 0.5452587604522705, "lr": 0.0001, "grad_norm": 0.14935913322646352, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90252392795, "tokens_per_second": 152336.71211892692, "data_wait_s": 1.3835999197908677e-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 sha256:3d6af850f5103e5d6e67c9d5304e2fd643ce050cb7862810220ab62f0113339b
-size 258792
+oid sha256:d80be094d6a363647af71f75dfc6e8a6ee1d0569c2a86ed811b45cf850880dfd
+size 260910