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": 77397.58794160942,
-      "required_pages_per_second_upper_estimate": 0.005354544720946845
+      "required_token_positions_per_second": 77028.38987016947,
+      "required_pages_per_second_upper_estimate": 0.005329002741707079
     },
     {
       "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": 9674.698492701178,
-      "required_pages_per_second_upper_estimate": 0.004050594228215782
+      "required_token_positions_per_second": 9628.548733771184,
+      "required_pages_per_second_upper_estimate": 0.004031272287868347
     },
     {
       "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": 19349.396985402356,
-      "required_pages_per_second_upper_estimate": 0.004029969114146103
+      "required_token_positions_per_second": 19257.097467542368,
+      "required_pages_per_second_upper_estimate": 0.004010745558677839
     },
     {
       "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": 29024.095478103533,
-      "required_pages_per_second_upper_estimate": 0.0017399502066845624
+      "required_token_positions_per_second": 28885.646201313553,
+      "required_pages_per_second_upper_estimate": 0.0017316503839408083
     },
     {
       "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": 29024.095478103533,
-      "required_pages_per_second_upper_estimate": 0.0022564957599184366
+      "required_token_positions_per_second": 28885.646201313553,
+      "required_pages_per_second_upper_estimate": 0.002245731937622026
     },
     {
       "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": 29024.095478103533,
-      "required_pages_per_second_upper_estimate": 0.37228993314738823
+      "required_token_positions_per_second": 28885.646201313553,
+      "required_pages_per_second_upper_estimate": 0.3705140544799779
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.08564852803134215,
-    "producer_p95_seconds": 0.19516599400049017,
-    "gpu_update_seconds": 0.9525878255485449,
-    "producer_batches_per_second": 11.675623889695581,
-    "gpu_batches_per_second": 1.049771971864277,
-    "producer_headroom": 11.122057173007757,
+    "producer_batch_seconds": 0.07203421175015023,
+    "producer_p95_seconds": 0.17440752599941334,
+    "gpu_update_seconds": 0.9571535913481738,
+    "producer_batches_per_second": 13.882292534393068,
+    "gpu_batches_per_second": 1.0447644025359357,
+    "producer_headroom": 13.287486155440266,
     "sustainable": true,
-    "buffer_seconds": 60.01303300955833,
+    "buffer_seconds": 60.30067625493495,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 292191.7222341402,
-    "upload_bytes_per_second": 4465411.89282262,
+    "generation_bytes_per_second": 290553.7712133244,
+    "upload_bytes_per_second": 4461358.525840384,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.69754171627192,
+    "estimated_full_shards_per_hour": 62.34607555675315,
     "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": [
@@ -502,6 +502,12 @@
       "bytes": 7751245,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.385997141000189,
+      "bytes": 5854774,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -82,3 +82,4 @@
 {"step": 533200, "loss": 0.5814388394355774, "lr": 0.0003, "grad_norm": 0.15108192271901014, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78043485668, "tokens_per_second": 153120.8393118724, "data_wait_s": 2.0298000890761614e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 533300, "loss": 0.546099066734314, "lr": 0.0003, "grad_norm": 0.16129710233637432, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78058151537, "tokens_per_second": 153524.2972613156, "data_wait_s": 8.629000149085186e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 533400, "loss": 0.6067091822624207, "lr": 0.0003, "grad_norm": 0.1560423516270421, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78072913636, "tokens_per_second": 154132.26853303952, "data_wait_s": 1.4012999599799514e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 533500, "loss": 0.6668593287467957, "lr": 0.0003, "grad_norm": 0.2902200800299181, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78087696348, "tokens_per_second": 154199.5834391522, "data_wait_s": 8.674000127939507e-06, "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 sha256:6301b000468644d4284f3e5a4da4fbd16a545b0b455057555c9467236cf6499a
-size 178860
+oid sha256:0b2eb53254a55af05a705e59c6836b60936cb453d5c656dd9c6d64590a56caa5
+size 180978