Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76061.81730694261,
-      "required_pages_per_second_upper_estimate": 0.006592845547252273
+      "required_token_positions_per_second": 75805.50815081931,
+      "required_pages_per_second_upper_estimate": 0.006570629319209122
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9507.727163367827,
-      "required_pages_per_second_upper_estimate": 0.0037927745186563853
+      "required_token_positions_per_second": 9475.688518852414,
+      "required_pages_per_second_upper_estimate": 0.003779993824338764
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19015.454326735653,
-      "required_pages_per_second_upper_estimate": 0.00439889069792422
+      "required_token_positions_per_second": 18951.377037704828,
+      "required_pages_per_second_upper_estimate": 0.004384067544828709
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28523.181490103478,
-      "required_pages_per_second_upper_estimate": 0.0017369121031795512
+      "required_token_positions_per_second": 28427.065556557245,
+      "required_pages_per_second_upper_estimate": 0.001731059147107916
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28523.181490103478,
-      "required_pages_per_second_upper_estimate": 0.002195793468328017
+      "required_token_positions_per_second": 28427.065556557245,
+      "required_pages_per_second_upper_estimate": 0.002188394197697695
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28523.181490103478,
-      "required_pages_per_second_upper_estimate": 0.3541624531594606
+      "required_token_positions_per_second": 28427.065556557245,
+      "required_pages_per_second_upper_estimate": 0.35296901494415295
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.10084019580472159,
-    "producer_p95_seconds": 0.3984677810003632,
-    "gpu_update_seconds": 0.9693168347855188,
-    "producer_batches_per_second": 9.916680466750716,
-    "gpu_batches_per_second": 1.031654423108488,
-    "producer_headroom": 9.612405321610185,
+    "producer_batch_seconds": 0.15192274189064392,
+    "producer_p95_seconds": 0.40610709699831204,
+    "gpu_update_seconds": 0.9725942322464748,
+    "producer_batches_per_second": 6.582292996790525,
+    "gpu_batches_per_second": 1.028178007687979,
+    "producer_headroom": 6.4019002036348285,
     "sustainable": true,
-    "buffer_seconds": 60.09764375670216,
+    "buffer_seconds": 60.300842399281436,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 320349.04342927865,
-    "upload_bytes_per_second": 6148257.963873274,
+    "generation_bytes_per_second": 320050.37398955424,
+    "upload_bytes_per_second": 6138764.972522554,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.73944737585802,
+    "estimated_full_shards_per_hour": 68.67535986676188,
     "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": 1.1064680679992307,
-      "bytes": 7358899,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 1.2355712200005655,
       "bytes": 6983067,
@@ -562,6 +556,12 @@
       "bytes": 11924691,
       "pages": 42,
       "shards": 3
+    },
+    {
+      "seconds": 0.9983105119972606,
+      "bytes": 6041328,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -373,3 +373,4 @@
 {"step": 77300, "loss": 0.6141677498817444, "lr": 0.0003, "grad_norm": 0.13769632136882737, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11312831331, "tokens_per_second": 147915.06908929613, "data_wait_s": 8.301998605020344e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 77400, "loss": 0.6213783863931894, "lr": 0.0003, "grad_norm": 0.14776273378710894, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11327293537, "tokens_per_second": 147072.40065249059, "data_wait_s": 6.088004738558084e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 77500, "loss": 0.5651458688080311, "lr": 0.0003, "grad_norm": 0.14182728180587317, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11341835766, "tokens_per_second": 147444.22801311596, "data_wait_s": 1.221999991685152e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 77600, "loss": 0.5945443157106638, "lr": 0.0003, "grad_norm": 0.12480173409034499, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11356454331, "tokens_per_second": 149123.80973602796, "data_wait_s": 1.2174998119007796e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3459e59278fe3efaae60cd004aa6a24e96b4b11ff645d6f937c274de0da4abd1
-size 799896
+oid sha256:2d37b6afa4f11b17eeea6607d6af6b2893bc3fe63d0e3e4f4b2b0d00672365e4
+size 802014