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": 76679.98838055957,
-      "required_pages_per_second_upper_estimate": 0.005304899518253034
+      "required_token_positions_per_second": 76484.13575240434,
+      "required_pages_per_second_upper_estimate": 0.005291349978996548
     },
     {
       "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": 9584.998547569947,
-      "required_pages_per_second_upper_estimate": 0.004013038734337192
+      "required_token_positions_per_second": 9560.516969050543,
+      "required_pages_per_second_upper_estimate": 0.0040027888086446115
     },
     {
       "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": 19169.997095139894,
-      "required_pages_per_second_upper_estimate": 0.00399260484809769
+      "required_token_positions_per_second": 19121.033938101085,
+      "required_pages_per_second_upper_estimate": 0.00398240711373179
     },
     {
       "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": 28754.99564270984,
-      "required_pages_per_second_upper_estimate": 0.0017238180824443677
+      "required_token_positions_per_second": 28681.550907151624,
+      "required_pages_per_second_upper_estimate": 0.0017194151826912707
     },
     {
       "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": 28754.99564270984,
-      "required_pages_per_second_upper_estimate": 0.0022355744313616614
+      "required_token_positions_per_second": 28681.550907151624,
+      "required_pages_per_second_upper_estimate": 0.0022298644262212608
     },
     {
       "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": 28754.99564270984,
-      "required_pages_per_second_upper_estimate": 0.3688382093958497
+      "required_token_positions_per_second": 28681.550907151624,
+      "required_pages_per_second_upper_estimate": 0.367896139186922
     }
   ],
   "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": 0.08222956352342692,
-    "producer_p95_seconds": 0.20917661400017096,
-    "gpu_update_seconds": 0.9615024931158181,
-    "producer_batches_per_second": 12.161076347135218,
-    "gpu_batches_per_second": 1.0400389049012528,
-    "producer_headroom": 11.692905226742319,
+    "producer_batch_seconds": 0.12936961300005123,
+    "producer_p95_seconds": 0.2445798280004965,
+    "gpu_update_seconds": 0.9639646088003591,
+    "producer_batches_per_second": 7.729790457049631,
+    "gpu_batches_per_second": 1.0373824836209355,
+    "producer_headroom": 7.451244434038597,
     "sustainable": true,
-    "buffer_seconds": 60.574657066296545,
+    "buffer_seconds": 60.72977035442262,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 518812.3566118237,
-    "upload_bytes_per_second": 3893918.8439946305,
+    "generation_bytes_per_second": 529883.9151757492,
+    "upload_bytes_per_second": 3890490.478963463,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 111.32505439535173,
+    "estimated_full_shards_per_hour": 113.7007531304775,
     "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": 1.540582020000329,
-      "bytes": 7589375,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.573228470999311,
       "bytes": 6089737,
@@ -562,6 +556,12 @@
       "bytes": 1261477,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.6420892379992438,
+      "bytes": 7607765,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -141,3 +141,4 @@
 {"step": 814100, "loss": 0.5360915660858154, "lr": 3e-05, "grad_norm": 0.27407967578988174, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119160974846, "tokens_per_second": 152299.70054497197, "data_wait_s": 1.3650000255438499e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 814200, "loss": 0.5107744932174683, "lr": 3e-05, "grad_norm": 0.276046562972792, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119175601574, "tokens_per_second": 151246.73205593054, "data_wait_s": 9.609000699128956e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 814300, "loss": 0.5602394938468933, "lr": 3e-05, "grad_norm": 0.27729984187245255, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119190194816, "tokens_per_second": 151159.33371674424, "data_wait_s": 7.97600114310626e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 814400, "loss": 0.5305436253547668, "lr": 3e-05, "grad_norm": 0.27287402219806217, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119204952198, "tokens_per_second": 129982.8005996017, "data_wait_s": 8.906999937607907e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e0a54fde3ed5e8c26d10b5723b0959cb1a39dcbde18573b39a00210701a5532b
-size 304605
+oid sha256:47fe805a85780e5ad8f9467b6c2606362439cfb194d5cbfaa4912aee0e062256
+size 306723