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": 75587.75343482329,
-      "required_pages_per_second_upper_estimate": 0.00655175489232087
+      "required_token_positions_per_second": 75365.73736685616,
+      "required_pages_per_second_upper_estimate": 0.006532511102244059
     },
     {
       "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": 9448.469179352911,
-      "required_pages_per_second_upper_estimate": 0.0037691356228470203
+      "required_token_positions_per_second": 9420.71717085702,
+      "required_pages_per_second_upper_estimate": 0.0037580649317285066
     },
     {
       "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": 18896.938358705822,
-      "required_pages_per_second_upper_estimate": 0.004371474114530317
+      "required_token_positions_per_second": 18841.43434171404,
+      "required_pages_per_second_upper_estimate": 0.004358634237036604
     },
     {
       "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": 28345.40753805873,
-      "required_pages_per_second_upper_estimate": 0.001726086602207869
+      "required_token_positions_per_second": 28262.15151257106,
+      "required_pages_per_second_upper_estimate": 0.0017210167470662772
     },
     {
       "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": 28345.40753805873,
-      "required_pages_per_second_upper_estimate": 0.0021821079373898168
+      "required_token_positions_per_second": 28262.15151257106,
+      "required_pages_per_second_upper_estimate": 0.002175698658080347
     },
     {
       "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": 28345.40753805873,
-      "required_pages_per_second_upper_estimate": 0.35195509564620897
+      "required_token_positions_per_second": 28262.15151257106,
+      "required_pages_per_second_upper_estimate": 0.350921334449645
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.05527458062486801,
-    "producer_p95_seconds": 0.18658741200124496,
-    "gpu_update_seconds": 0.9753961012159611,
-    "producer_batches_per_second": 18.091498636356917,
-    "gpu_batches_per_second": 1.0252245203290917,
-    "producer_headroom": 17.646377235056413,
+    "producer_batch_seconds": 0.11490734128896918,
+    "producer_p95_seconds": 0.20288106900079583,
+    "gpu_update_seconds": 0.978269470662986,
+    "producer_batches_per_second": 8.702664153417302,
+    "gpu_batches_per_second": 1.0222132346850066,
+    "producer_headroom": 8.513550654721287,
     "sustainable": true,
-    "buffer_seconds": 60.47455827538959,
+    "buffer_seconds": 60.65270718110513,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 274680.0010494453,
-    "upload_bytes_per_second": 5899712.91558434,
+    "generation_bytes_per_second": 315534.8649662977,
+    "upload_bytes_per_second": 5853113.14753,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 58.939934002042,
+    "estimated_full_shards_per_hour": 67.70643674604128,
     "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": 1.0488426049996633,
-      "bytes": 5758950,
-      "pages": 26,
-      "shards": 2
-    },
     {
       "seconds": 1.0318703240000104,
       "bytes": 8301935,
@@ -562,6 +556,12 @@
       "bytes": 5878114,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 0.8248095879989705,
+      "bytes": 1051965,
+      "pages": 33,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -115,3 +115,4 @@
 {"step": 366500, "loss": 0.6233857721090317, "lr": 0.0003, "grad_norm": 0.1690511630703369, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53645461592, "tokens_per_second": 150633.3828324391, "data_wait_s": 8.834998880047351e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 366600, "loss": 0.6111076474189758, "lr": 0.0003, "grad_norm": 0.1749200668043921, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53660217163, "tokens_per_second": 150523.07928463042, "data_wait_s": 1.1516000085975975e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 366700, "loss": 0.6004631537944078, "lr": 0.0003, "grad_norm": 0.16477317280781276, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53674931778, "tokens_per_second": 149537.63814292118, "data_wait_s": 1.3441000191960484e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 366800, "loss": 0.5651937630027533, "lr": 0.0003, "grad_norm": 0.1773074711262574, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53689494372, "tokens_per_second": 149013.9721889122, "data_wait_s": 1.0245999874314293e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:24eaaf0eed318cba6efa668802327d905117975f63a2836d85594e3f39c2e79e
-size 249537
+oid sha256:2a258ee451feace554f94c0983927634c8c049ff48a3cefeaf9455d7d53776e9
+size 251655