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": 76442.5030577095,
-      "required_pages_per_second_upper_estimate": 0.0052884697325763985
+      "required_token_positions_per_second": 76897.88463404337,
+      "required_pages_per_second_upper_estimate": 0.005319974086658007
     },
     {
       "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": 9555.312882213688,
-      "required_pages_per_second_upper_estimate": 0.004000609966159711
+      "required_token_positions_per_second": 9612.23557925542,
+      "required_pages_per_second_upper_estimate": 0.004024442310730001
     },
     {
       "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": 19110.625764427376,
-      "required_pages_per_second_upper_estimate": 0.003980239365637554
+      "required_token_positions_per_second": 19224.47115851084,
+      "required_pages_per_second_upper_estimate": 0.004003950358920202
     },
     {
       "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": 28665.938646641065,
-      "required_pages_per_second_upper_estimate": 0.001718479251512204
+      "required_token_positions_per_second": 28836.706737766264,
+      "required_pages_per_second_upper_estimate": 0.0017287165378273824
     },
     {
       "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": 28665.938646641065,
-      "required_pages_per_second_upper_estimate": 0.0022286506416376414
+      "required_token_positions_per_second": 28836.706737766264,
+      "required_pages_per_second_upper_estimate": 0.0022419271095931747
     },
     {
       "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": 28665.938646641065,
-      "required_pages_per_second_upper_estimate": 0.36769588187223184
+      "required_token_positions_per_second": 28836.706737766264,
+      "required_pages_per_second_upper_estimate": 0.36988631158869517
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.08313471499209868,
-    "producer_p95_seconds": 0.20901094299915712,
-    "gpu_update_seconds": 0.9644896105029395,
-    "producer_batches_per_second": 12.028669372295825,
-    "gpu_batches_per_second": 1.036817804059645,
-    "producer_headroom": 11.601526637754239,
+    "producer_batch_seconds": 0.1188022856404416,
+    "producer_p95_seconds": 0.21895740300169564,
+    "gpu_update_seconds": 0.9587779995622918,
+    "producer_batches_per_second": 8.417346472832413,
+    "gpu_batches_per_second": 1.0429943119851803,
+    "producer_headroom": 8.070366612844975,
     "sustainable": true,
-    "buffer_seconds": 60.76284546168519,
+    "buffer_seconds": 60.40301397242438,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 338166.24984439893,
-    "upload_bytes_per_second": 4601510.299395956,
+    "generation_bytes_per_second": 325042.3849593266,
+    "upload_bytes_per_second": 4593402.590772988,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.56260510920502,
+    "estimated_full_shards_per_hour": 69.74652921280716,
     "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": 1.0025934540008166,
-      "bytes": 6106895,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.0866103590014973,
       "bytes": 4773621,
@@ -562,6 +556,12 @@
       "bytes": 11886724,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 1.1407382979996328,
+      "bytes": 5980776,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -184,3 +184,4 @@
 {"step": 458600, "loss": 0.5902373790740967, "lr": 0.0003, "grad_norm": 0.1600202144644638, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67124795481, "tokens_per_second": 151802.97186745255, "data_wait_s": 1.3360000593820587e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 458700, "loss": 0.6024937629699707, "lr": 0.0003, "grad_norm": 0.16354908680616428, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67139479315, "tokens_per_second": 152337.7971899356, "data_wait_s": 1.2299999070819467e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 458800, "loss": 0.5767513513565063, "lr": 0.0003, "grad_norm": 0.15123196371747163, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67154198469, "tokens_per_second": 153323.8081858713, "data_wait_s": 1.034600063576363e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 458900, "loss": 0.5333598256111145, "lr": 0.0003, "grad_norm": 0.1435300061299573, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67168875836, "tokens_per_second": 153136.1474479134, "data_wait_s": 8.0740028352011e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:81885de40d65616f310a99e787cba9300bf64763ae6d7a37e17b7617675d8a81
-size 396462
+oid sha256:51939aee83be51357b3fa98dd7f38a58100ae5d3f1588395bcb49b174ec19d5d
+size 398580