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": 77684.31378943461,
-      "required_pages_per_second_upper_estimate": 0.005374381080395018
+      "required_token_positions_per_second": 77652.63219599545,
+      "required_pages_per_second_upper_estimate": 0.005372189274249472
     },
     {
       "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": 9710.539223679327,
-      "required_pages_per_second_upper_estimate": 0.004065599993836762
+      "required_token_positions_per_second": 9706.579024499431,
+      "required_pages_per_second_upper_estimate": 0.004063941941138502
     },
     {
       "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": 19421.078447358654,
-      "required_pages_per_second_upper_estimate": 0.004044898472304326
+      "required_token_positions_per_second": 19413.158048998863,
+      "required_pages_per_second_upper_estimate": 0.004043248862200932
     },
     {
       "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": 29131.617671037977,
-      "required_pages_per_second_upper_estimate": 0.0017463959979740998
+      "required_token_positions_per_second": 29119.73707349829,
+      "required_pages_per_second_upper_estimate": 0.0017456837742922178
     },
     {
       "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": 29131.617671037977,
-      "required_pages_per_second_upper_estimate": 0.0022648551374789445
+      "required_token_positions_per_second": 29119.73707349829,
+      "required_pages_per_second_upper_estimate": 0.00226393147327746
     },
     {
       "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": 29131.617671037977,
-      "required_pages_per_second_upper_estimate": 0.3736691123900152
+      "required_token_positions_per_second": 29119.73707349829,
+      "required_pages_per_second_upper_estimate": 0.3735167208411679
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.12113308810947387,
-    "producer_p95_seconds": 0.22835545599991747,
-    "gpu_update_seconds": 0.9490719091609884,
-    "producer_batches_per_second": 8.255382700193785,
-    "gpu_batches_per_second": 1.0536609400693713,
-    "producer_headroom": 7.83495182012751,
+    "producer_batch_seconds": 0.061487864132772074,
+    "producer_p95_seconds": 0.2082157000004372,
+    "gpu_update_seconds": 0.9494591221828815,
+    "producer_batches_per_second": 16.263371871897817,
+    "gpu_batches_per_second": 1.0532312309569696,
+    "producer_headroom": 15.441406781225869,
     "sustainable": true,
-    "buffer_seconds": 60.740602186303256,
+    "buffer_seconds": 60.76538381970442,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 397114.56439943874,
-    "upload_bytes_per_second": 5785240.091619257,
+    "generation_bytes_per_second": 397060.55581649387,
+    "upload_bytes_per_second": 5643217.795535816,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.21154116618511,
+    "estimated_full_shards_per_hour": 85.19995218154061,
     "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": 1.0915457699993567,
-      "bytes": 13214903,
-      "pages": 39,
-      "shards": 4
-    },
     {
       "seconds": 1.4333472569996957,
       "bytes": 6210705,
@@ -562,6 +556,12 @@
       "bytes": 5544847,
       "pages": 18,
       "shards": 2
+    },
+    {
+      "seconds": 1.068828384999506,
+      "bytes": 2455135,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -150,3 +150,4 @@
 {"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}
 {"step": 540200, "loss": 0.588968813419342, "lr": 0.0003, "grad_norm": 0.1545030960327682, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79067116593, "tokens_per_second": 151242.03180545435, "data_wait_s": 1.235899981111288e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 540300, "loss": 0.6125973463058472, "lr": 0.0003, "grad_norm": 0.18708997728325327, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79081606092, "tokens_per_second": 151298.96701039077, "data_wait_s": 1.1809999705292284e-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 sha256:5ec2e4cdece83158aa97afa7641d4b07268c855865d62cfb40d509f2e4cf59d5
-size 324450
+oid sha256:f4ffd5bc968071bb950f24189bb1f0fcd05a385686c9c23b8b1b2aaa297f89c6
+size 326568