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": 77237.00682912544,
-      "required_pages_per_second_upper_estimate": 0.005343435346985688
+      "required_token_positions_per_second": 77326.14539960928,
+      "required_pages_per_second_upper_estimate": 0.005349602160121738
     },
     {
       "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": 9654.62585364068,
-      "required_pages_per_second_upper_estimate": 0.004042190233405519
+      "required_token_positions_per_second": 9665.76817495116,
+      "required_pages_per_second_upper_estimate": 0.004046855290660089
     },
     {
       "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": 19309.25170728136,
-      "required_pages_per_second_upper_estimate": 0.004021607911415677
+      "required_token_positions_per_second": 19331.53634990232,
+      "required_pages_per_second_upper_estimate": 0.004026249214788077
     },
     {
       "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": 28963.87756092204,
-      "required_pages_per_second_upper_estimate": 0.0017363402344969675
+      "required_token_positions_per_second": 28997.30452485348,
+      "required_pages_per_second_upper_estimate": 0.0017383441299446388
     },
     {
       "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": 28963.87756092204,
-      "required_pages_per_second_upper_estimate": 0.0022518140817282006
+      "required_token_positions_per_second": 28997.30452485348,
+      "required_pages_per_second_upper_estimate": 0.0022544128811441944
     },
     {
       "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": 28963.87756092204,
-      "required_pages_per_second_upper_estimate": 0.371517522362746
+      "required_token_positions_per_second": 28997.30452485348,
+      "required_pages_per_second_upper_estimate": 0.37194628756498094
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.20105687201572664,
-    "producer_p95_seconds": 0.32645556600618875,
-    "gpu_update_seconds": 0.9545683219329232,
-    "producer_batches_per_second": 4.973717087977874,
-    "gpu_batches_per_second": 1.0475939511328862,
-    "producer_headroom": 4.747752774440144,
+    "producer_batch_seconds": 0.23314612662522904,
+    "producer_p95_seconds": 0.3290378910023719,
+    "gpu_update_seconds": 0.9534679327281267,
+    "producer_batches_per_second": 4.289155537237172,
+    "gpu_batches_per_second": 1.0488029703723045,
+    "producer_headroom": 4.0895722632389235,
     "sustainable": true,
-    "buffer_seconds": 60.13780428177416,
+    "buffer_seconds": 60.068479761871984,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 528761.5025732169,
-    "upload_bytes_per_second": 3659528.1670097695,
+    "generation_bytes_per_second": 518569.6749402506,
+    "upload_bytes_per_second": 3740249.714373933,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 113.4599095144022,
+    "estimated_full_shards_per_hour": 111.27298055797232,
     "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": 0.976799342999584,
-      "bytes": 895629,
-      "pages": 27,
-      "shards": 1
-    },
     {
       "seconds": 1.1010828580037924,
       "bytes": 6038977,
@@ -562,6 +556,12 @@
       "bytes": 6050652,
       "pages": 32,
       "shards": 2
+    },
+    {
+      "seconds": 1.168204486995819,
+      "bytes": 10922588,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -402,3 +402,4 @@
 {"step": 525200, "loss": 0.6116625070571899, "lr": 0.0003, "grad_norm": 0.16057935794230394, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76871945986, "tokens_per_second": 155239.25487834294, "data_wait_s": 1.3882003258913755e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 525300, "loss": 0.5752341151237488, "lr": 0.0003, "grad_norm": 0.1445069745388966, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76886579996, "tokens_per_second": 152337.33543575293, "data_wait_s": 1.4636993000749499e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 525400, "loss": 0.5551020503044128, "lr": 0.0003, "grad_norm": 0.18036624674072047, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76901085601, "tokens_per_second": 151813.70222665436, "data_wait_s": 1.6175996279343963e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 525500, "loss": 0.5803568959236145, "lr": 0.0003, "grad_norm": 0.16922392638931597, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76915582593, "tokens_per_second": 151008.58769293266, "data_wait_s": 1.3894998119212687e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1d19f3793bba6fe9add75e880bf0d18b57c59c0999254a3dc4ff31b822025da6
-size 862101
+oid sha256:32b45bd4229480ef6bf0f5efd6df6aa579f012f7e4798481638148433ddbd054
+size 864219