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": 75033.58321646866,
-      "required_pages_per_second_upper_estimate": 0.006503720822325467
+      "required_token_positions_per_second": 75544.88242264869,
+      "required_pages_per_second_upper_estimate": 0.0065480389416411626
     },
     {
       "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": 9379.197902058582,
-      "required_pages_per_second_upper_estimate": 0.003741502274636422
+      "required_token_positions_per_second": 9443.110302831086,
+      "required_pages_per_second_upper_estimate": 0.003766997886880121
     },
     {
       "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": 18758.395804117165,
-      "required_pages_per_second_upper_estimate": 0.0043394247327919145
+      "required_token_positions_per_second": 18886.220605662173,
+      "required_pages_per_second_upper_estimate": 0.00436899475098968
     },
     {
       "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": 28137.593706175747,
-      "required_pages_per_second_upper_estimate": 0.0017134318301611072
+      "required_token_positions_per_second": 28329.33090849326,
+      "required_pages_per_second_upper_estimate": 0.0017251076198148885
     },
     {
       "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": 28137.593706175747,
-      "required_pages_per_second_upper_estimate": 0.0021661098533459597
+      "required_token_positions_per_second": 28329.33090849326,
+      "required_pages_per_second_upper_estimate": 0.002180870313942908
     },
     {
       "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": 28137.593706175747,
-      "required_pages_per_second_upper_estimate": 0.3493747433623769
+      "required_token_positions_per_second": 28329.33090849326,
+      "required_pages_per_second_upper_estimate": 0.35175547771202376
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.14808333250027772,
-    "producer_p95_seconds": 0.3791705350013217,
-    "gpu_update_seconds": 0.9826000150798863,
-    "producer_batches_per_second": 6.752954455547012,
-    "gpu_batches_per_second": 1.0177081056921204,
-    "producer_headroom": 6.635453149854278,
+    "producer_batch_seconds": 0.09331862435135463,
+    "producer_p95_seconds": 0.37257002700062003,
+    "gpu_update_seconds": 0.9759496293543243,
+    "producer_batches_per_second": 10.715974511528296,
+    "gpu_batches_per_second": 1.0246430450120536,
+    "producer_headroom": 10.458251352696427,
     "sustainable": true,
-    "buffer_seconds": 60.92120093495295,
+    "buffer_seconds": 60.508877019968104,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 284891.3441172064,
+    "generation_bytes_per_second": 289292.7052835849,
     "upload_bytes_per_second": 5290840.456530075,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.13105051648277,
+    "estimated_full_shards_per_hour": 62.07548016434345,
     "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": [

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -353,3 +353,4 @@
 {"step": 272400, "loss": 0.6329742148518562, "lr": 0.0003, "grad_norm": 0.142750450724264, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39870833835, "tokens_per_second": 148554.5394107169, "data_wait_s": 1.657800021348521e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 272500, "loss": 0.5857648644596338, "lr": 0.0003, "grad_norm": 0.16595507092760214, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39885375281, "tokens_per_second": 136249.45000694657, "data_wait_s": 1.1500997061375529e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 272600, "loss": 0.5896847527474165, "lr": 0.0003, "grad_norm": 0.1604071163043989, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39899861492, "tokens_per_second": 147266.40900139642, "data_wait_s": 1.03089987533167e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 272700, "loss": 0.6070579923689365, "lr": 0.0003, "grad_norm": 0.14527561808493308, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39914524766, "tokens_per_second": 149764.96514063634, "data_wait_s": 1.1429001460783184e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7c12f3a19400d891fa3b65a32abdc16d7e7cd676e2e1250abf8910cfa36d8555
-size 757536
+oid sha256:456d085f6a7819e90ab58fc67869076f15dc53f88ff7c52cacb617cad311281f
+size 759654