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": 76567.33905992791,
-      "required_pages_per_second_upper_estimate": 0.0052971061768693
+      "required_token_positions_per_second": 76809.51770117912,
+      "required_pages_per_second_upper_estimate": 0.005313860657202926
     },
     {
       "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": 9570.917382490989,
-      "required_pages_per_second_upper_estimate": 0.004007143244566796
+      "required_token_positions_per_second": 9601.18971264739,
+      "required_pages_per_second_upper_estimate": 0.004019817637045143
     },
     {
       "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": 19141.834764981977,
-      "required_pages_per_second_upper_estimate": 0.003986739377416386
+      "required_token_positions_per_second": 19202.37942529478,
+      "required_pages_per_second_upper_estimate": 0.003999349233489479
     },
     {
       "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": 28712.752147472966,
-      "required_pages_per_second_upper_estimate": 0.0017212856494069997
+      "required_token_positions_per_second": 28803.56913794217,
+      "required_pages_per_second_upper_estimate": 0.0017267299893161132
     },
     {
       "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": 28712.752147472966,
-      "required_pages_per_second_upper_estimate": 0.002232290185416493
+      "required_token_positions_per_second": 28803.56913794217,
+      "required_pages_per_second_upper_estimate": 0.002239350806963749
     },
     {
       "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": 28712.752147472966,
-      "required_pages_per_second_upper_estimate": 0.3682963551964824
+      "required_token_positions_per_second": 28803.56913794217,
+      "required_pages_per_second_upper_estimate": 0.36946125803853425
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.08247454904690699,
-    "producer_p95_seconds": 0.2957663870001852,
-    "gpu_update_seconds": 0.9629170989250964,
-    "producer_batches_per_second": 12.124952625460917,
-    "gpu_batches_per_second": 1.0385110007043175,
-    "producer_headroom": 11.675324206713055,
+    "producer_batch_seconds": 0.13177555471872893,
+    "producer_p95_seconds": 0.3089953119997517,
+    "gpu_update_seconds": 0.9598810434773526,
+    "producer_batches_per_second": 7.588660902505558,
+    "gpu_batches_per_second": 1.041795758750802,
+    "producer_headroom": 7.284211745692823,
     "sustainable": true,
-    "buffer_seconds": 60.66377723228107,
+    "buffer_seconds": 60.47250573907321,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 334995.3667352742,
-    "upload_bytes_per_second": 5548298.119672106,
+    "generation_bytes_per_second": 350547.27153319644,
+    "upload_bytes_per_second": 5533148.995422451,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.88220740836782,
+    "estimated_full_shards_per_hour": 75.21928414818687,
     "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": 1.4597207000006165,
-      "bytes": 10914765,
-      "pages": 43,
-      "shards": 3
-    },
     {
       "seconds": 1.293525116998353,
       "bytes": 5848497,
@@ -562,6 +556,12 @@
       "bytes": 5975966,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.119130936000147,
+      "bytes": 7848522,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -302,3 +302,4 @@
 {"step": 675200, "loss": 0.5657216310501099, "lr": 0.0001, "grad_norm": 0.2053365528525142, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98831507844, "tokens_per_second": 152895.0128957328, "data_wait_s": 1.4689998351968825e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 675300, "loss": 0.5566164255142212, "lr": 0.0001, "grad_norm": 0.2146848906602766, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98846120568, "tokens_per_second": 152344.26664891464, "data_wait_s": 1.751099989633076e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 675400, "loss": 0.5043987035751343, "lr": 0.0001, "grad_norm": 0.20687713845321776, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98860863971, "tokens_per_second": 154246.7165421657, "data_wait_s": 1.3676002708962187e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 675500, "loss": 0.5495113134384155, "lr": 0.0001, "grad_norm": 0.1809195087755194, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98875539308, "tokens_per_second": 152190.1896053792, "data_wait_s": 1.3176999345887452e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9dbf6d650e53c43755bdbb348549baf840187a650ee922161f19034b015e0bbf
-size 648735
+oid sha256:17cc081941d267fe7e131e4f3b9362d5872273d55b65ad9f8c9ab008646aa00c
+size 650853