Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75200.65572172558,
-      "required_pages_per_second_upper_estimate": 0.0065182022436397445
+      "required_token_positions_per_second": 75836.43573221067,
+      "required_pages_per_second_upper_estimate": 0.006573310043578887
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9400.081965215697,
-      "required_pages_per_second_upper_estimate": 0.0037498332396743647
+      "required_token_positions_per_second": 9479.554466526333,
+      "required_pages_per_second_upper_estimate": 0.0037815360086669593
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18800.163930431394,
-      "required_pages_per_second_upper_estimate": 0.004349087053720795
+      "required_token_positions_per_second": 18959.108933052667,
+      "required_pages_per_second_upper_estimate": 0.0043858561827407215
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28200.24589564709,
-      "required_pages_per_second_upper_estimate": 0.001717247019789278
+      "required_token_positions_per_second": 28438.663399579,
+      "required_pages_per_second_upper_estimate": 0.0017317653948987612
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28200.24589564709,
-      "required_pages_per_second_upper_estimate": 0.0021709329923238242
+      "required_token_positions_per_second": 28438.663399579,
+      "required_pages_per_second_upper_estimate": 0.0021892870317583944
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28200.24589564709,
-      "required_pages_per_second_upper_estimate": 0.3501526738722214
+      "required_token_positions_per_second": 28438.663399579,
+      "required_pages_per_second_upper_estimate": 0.35311302133899947
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.11146160417996498,
-    "producer_p95_seconds": 0.2996972929977346,
-    "gpu_update_seconds": 0.9804169829692041,
-    "producer_batches_per_second": 8.971699334107988,
-    "gpu_batches_per_second": 1.0199741715728836,
-    "producer_headroom": 8.796006393252972,
+    "producer_batch_seconds": 0.4964295274923245,
+    "producer_p95_seconds": 0.325202015999821,
+    "gpu_update_seconds": 0.9721975893005329,
+    "producer_batches_per_second": 2.014384609737908,
+    "gpu_batches_per_second": 1.028597489857458,
+    "producer_headroom": 1.9583798615112886,
     "sustainable": true,
-    "buffer_seconds": 60.78585294409066,
+    "buffer_seconds": 60.27625053663304,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 316182.76767323195,
-    "upload_bytes_per_second": 5291326.802045232,
+    "generation_bytes_per_second": 294248.4065939075,
+    "upload_bytes_per_second": 5236919.2006158205,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.84546158454627,
+    "estimated_full_shards_per_hour": 63.138858302716436,
     "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": 1.2984945649986912,
-      "bytes": 7220956,
-      "pages": 28,
-      "shards": 3
-    },
     {
       "seconds": 1.4053506099990045,
       "bytes": 11055172,
@@ -562,6 +556,12 @@
       "bytes": 7262597,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 0.8794477230003395,
+      "bytes": 663465,
+      "pages": 19,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -189,3 +189,4 @@
 {"step": 173900, "loss": 0.5828830786049366, "lr": 0.0003, "grad_norm": 0.14228330439659898, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25451268564, "tokens_per_second": 148178.62506568458, "data_wait_s": 1.5075002011144534e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 174000, "loss": 0.5894587468355894, "lr": 0.0003, "grad_norm": 0.15220324222693843, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25465872070, "tokens_per_second": 148222.33531019682, "data_wait_s": 1.2658998457482085e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 174100, "loss": 0.5944762509316206, "lr": 0.0003, "grad_norm": 0.15662383173064823, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25480470463, "tokens_per_second": 127230.22103473838, "data_wait_s": 1.2153999705333263e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 174200, "loss": 0.6223274301737547, "lr": 0.0003, "grad_norm": 0.1410694148798424, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25495207625, "tokens_per_second": 149775.84034037788, "data_wait_s": 1.4260996977100149e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:349f27407c4540251c3ea5dc73dcf37dda5bab4732237fd978c163d8557ffc4e
-size 407052
+oid sha256:59ed60750fadda9672f26b56772db769917915ce3b04466eeb9450903d2496fd
+size 409170