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": 75723.45563191624,
-      "required_pages_per_second_upper_estimate": 0.0065635172148308125
+      "required_token_positions_per_second": 75108.72568282984,
+      "required_pages_per_second_upper_estimate": 0.006510233980862815
     },
     {
       "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": 9465.43195398953,
-      "required_pages_per_second_upper_estimate": 0.0037759023272656493
+      "required_token_positions_per_second": 9388.59071035373,
+      "required_pages_per_second_upper_estimate": 0.0037452492063003552
     },
     {
       "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": 18930.86390797906,
-      "required_pages_per_second_upper_estimate": 0.004379322193285406
+      "required_token_positions_per_second": 18777.18142070746,
+      "required_pages_per_second_upper_estimate": 0.00434377045457452
     },
     {
       "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": 28396.29586196859,
-      "required_pages_per_second_upper_estimate": 0.001729185434143578
+      "required_token_positions_per_second": 28165.772131061192,
+      "required_pages_per_second_upper_estimate": 0.0017151477483958613
     },
     {
       "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": 28396.29586196859,
-      "required_pages_per_second_upper_estimate": 0.002186025461432294
+      "required_token_positions_per_second": 28165.772131061192,
+      "required_pages_per_second_upper_estimate": 0.002168279106496514
     },
     {
       "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": 28396.29586196859,
-      "required_pages_per_second_upper_estimate": 0.3525869583169051
+      "required_token_positions_per_second": 28165.772131061192,
+      "required_pages_per_second_upper_estimate": 0.3497246250923326
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.12726434982027968,
-    "producer_p95_seconds": 0.3987374990028911,
-    "gpu_update_seconds": 0.9736481171485896,
-    "producer_batches_per_second": 7.857660070649645,
-    "gpu_batches_per_second": 1.027065099174211,
-    "producer_headroom": 7.65059593298168,
+    "producer_batch_seconds": 0.4212736248362603,
+    "producer_p95_seconds": 0.37903478399675805,
+    "gpu_update_seconds": 0.981616973656824,
+    "producer_batches_per_second": 2.373754113822525,
+    "gpu_batches_per_second": 1.0187272906199794,
+    "producer_headroom": 2.3301173294159034,
     "sustainable": true,
-    "buffer_seconds": 60.366183263212555,
+    "buffer_seconds": 60.86025236672309,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 334958.4564990026,
+    "generation_bytes_per_second": 327707.24065361626,
     "upload_bytes_per_second": 6104186.3553599045,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.87428733089025,
+    "estimated_full_shards_per_hour": 70.31834521013609,
     "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-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -375,3 +375,4 @@
 {"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}
 {"step": 77700, "loss": 0.6419476978480816, "lr": 0.0003, "grad_norm": 0.1387862551719045, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11371135539, "tokens_per_second": 150203.60170038426, "data_wait_s": 9.581002814229578e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 77800, "loss": 0.6180443204939365, "lr": 0.0003, "grad_norm": 0.14828734039842262, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11385783307, "tokens_per_second": 149977.69468966572, "data_wait_s": 1.0100004146806896e-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:ba184dcd49b43d0a63bd95ff675f5e8bbe106b599afabd77199b0814238c2735
-size 804132
+oid sha256:25923d5c6d25a8f6f840114ee6c1f0ad344145d5edfa2a0908c62d78c0e8fdbd
+size 806250