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": 76448.66075247723,
-      "required_pages_per_second_upper_estimate": 0.005288895736188249
+      "required_token_positions_per_second": 77368.1114725715,
+      "required_pages_per_second_upper_estimate": 0.00535250547042397
     },
     {
       "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": 9556.082594059653,
-      "required_pages_per_second_upper_estimate": 0.004000932228436205
+      "required_token_positions_per_second": 9671.013934071438,
+      "required_pages_per_second_upper_estimate": 0.004049051580459843
     },
     {
       "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": 19112.165188119307,
-      "required_pages_per_second_upper_estimate": 0.00398055998699525
+      "required_token_positions_per_second": 19342.027868142875,
+      "required_pages_per_second_upper_estimate": 0.004028434321357643
     },
     {
       "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": 28668.24778217896,
-      "required_pages_per_second_upper_estimate": 0.0017186176806618552
+      "required_token_positions_per_second": 29013.04180221431,
+      "required_pages_per_second_upper_estimate": 0.0017392875556929894
     },
     {
       "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": 28668.24778217896,
-      "required_pages_per_second_upper_estimate": 0.0022288301667688993
+      "required_token_positions_per_second": 29013.04180221431,
+      "required_pages_per_second_upper_estimate": 0.0022556363852380312
     },
     {
       "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": 28668.24778217896,
-      "required_pages_per_second_upper_estimate": 0.3677255009835554
+      "required_token_positions_per_second": 29013.04180221431,
+      "required_pages_per_second_upper_estimate": 0.3721481484615937
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,39 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.3263542902969334,
-    "producer_p95_seconds": 0.2078722389996983,
-    "gpu_update_seconds": 0.9644119239539579,
-    "producer_batches_per_second": 3.064154600480815,
-    "gpu_batches_per_second": 1.0369013231401534,
-    "producer_headroom": 2.9551072335420745,
+    "producer_batch_seconds": 0.08002629136724693,
+    "producer_p95_seconds": 0.21366449800007103,
+    "gpu_update_seconds": 0.9529507518887548,
+    "producer_batches_per_second": 12.49589332349442,
+    "gpu_batches_per_second": 1.0493721716657376,
+    "producer_headroom": 11.907970938145679,
     "sustainable": true,
-    "buffer_seconds": 60.757951209099346,
+    "buffer_seconds": 60.03589736899155,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 393197.65477415506,
-    "upload_bytes_per_second": 5782840.946028825,
+    "generation_bytes_per_second": 413907.74116544053,
+    "upload_bytes_per_second": 5785240.091619257,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.37106354158868,
+    "estimated_full_shards_per_hour": 88.81496597502148,
     "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": 1.4530193070004316,
-      "bytes": 7690650,
-      "pages": 38,
-      "shards": 3
-    },
-    {
-      "seconds": 1.0416680760008603,
-      "bytes": 5628205,
-      "pages": 24,
-      "shards": 2
-    },
     {
       "seconds": 1.0915457699993567,
       "bytes": 13214903,
@@ -562,6 +550,18 @@
       "bytes": 7770797,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.0592705330000172,
+      "bytes": 5143840,
+      "pages": 36,
+      "shards": 2
+    },
+    {
+      "seconds": 0.9495372719993611,
+      "bytes": 5544847,
+      "pages": 18,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -148,3 +148,4 @@
 {"step": 539800, "loss": 0.6331631541252136, "lr": 0.0003, "grad_norm": 0.18237055770298055, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79008423707, "tokens_per_second": 152120.11196001043, "data_wait_s": 8.068000170169398e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 539900, "loss": 0.6281508207321167, "lr": 0.0003, "grad_norm": 0.17473581720289452, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79023186415, "tokens_per_second": 154956.58886204657, "data_wait_s": 1.1304000508971512e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 540000, "loss": 0.5902459025382996, "lr": 0.0003, "grad_norm": 0.17200454621042746, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79037947186, "tokens_per_second": 154639.35389178383, "data_wait_s": 8.696999429957941e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 540100, "loss": 0.6129547953605652, "lr": 0.0003, "grad_norm": 0.17195902010706376, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79052648396, "tokens_per_second": 133635.71940093755, "data_wait_s": 1.3443999705486931e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f16b6772f2d213b939b8e5c43c7e7ac02ba32ef8b8db6c99fa7958404fc3e431
-size 320214
+oid sha256:c238cf5a0f3bdccb300e858992c7585e5ede84835431f8d83a90b5a943d23b85
+size 322332