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": 76142.98720157571,
-      "required_pages_per_second_upper_estimate": 0.006599881147995854
+      "required_token_positions_per_second": 75109.3984513659,
+      "required_pages_per_second_upper_estimate": 0.006510292294734404
     },
     {
       "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": 9517.873400196964,
-      "required_pages_per_second_upper_estimate": 0.0037968220042272874
+      "required_token_positions_per_second": 9388.674806420737,
+      "required_pages_per_second_upper_estimate": 0.0037452827534788322
     },
     {
       "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": 19035.74680039393,
-      "required_pages_per_second_upper_estimate": 0.004403585004569724
+      "required_token_positions_per_second": 18777.349612841474,
+      "required_pages_per_second_upper_estimate": 0.004343809362864643
     },
     {
       "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": 28553.620200590893,
-      "required_pages_per_second_upper_estimate": 0.0017387656609486632
+      "required_token_positions_per_second": 28166.024419262212,
+      "required_pages_per_second_upper_estimate": 0.0017151631114236494
     },
     {
       "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": 28553.620200590893,
-      "required_pages_per_second_upper_estimate": 0.002198136724521082
+      "required_token_positions_per_second": 28166.024419262212,
+      "required_pages_per_second_upper_estimate": 0.0021682985283406064
     },
     {
       "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": 28553.620200590893,
-      "required_pages_per_second_upper_estimate": 0.3545404000719035
+      "required_token_positions_per_second": 28166.024419262212,
+      "required_pages_per_second_upper_estimate": 0.34972775766743497
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.3122525599765922,
-    "producer_p95_seconds": 0.17674533999979758,
-    "gpu_update_seconds": 0.9682835243227004,
-    "producer_batches_per_second": 3.2025357937016254,
-    "gpu_batches_per_second": 1.0327553602644275,
-    "producer_headroom": 3.1009626450950063,
+    "producer_batch_seconds": 0.10289877708589756,
+    "producer_p95_seconds": 0.19374871000036364,
+    "gpu_update_seconds": 0.9816081811351429,
+    "producer_batches_per_second": 9.71828848038906,
+    "gpu_batches_per_second": 1.01873641562725,
+    "producer_headroom": 9.539551478981316,
     "sustainable": true,
-    "buffer_seconds": 60.03357850800742,
+    "buffer_seconds": 60.85970723037886,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 354383.0704747107,
-    "upload_bytes_per_second": 5034777.626014861,
+    "generation_bytes_per_second": 361005.0156141196,
+    "upload_bytes_per_second": 5098133.898308832,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.04235730820648,
+    "estimated_full_shards_per_hour": 77.4632725841302,
     "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": [
@@ -328,6 +328,12 @@
       "bytes": 4323709,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 1.0870202730002347,
+      "bytes": 7523830,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -58,3 +58,4 @@
 {"step": 160800, "loss": 0.5601533055305481, "lr": 0.0003, "grad_norm": 0.15310807497503692, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23534491326, "tokens_per_second": 150268.9764084363, "data_wait_s": 9.97200004348997e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 160900, "loss": 0.604211675003171, "lr": 0.0003, "grad_norm": 0.14165015113283677, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23548984978, "tokens_per_second": 148113.15121979677, "data_wait_s": 9.25599943002453e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 161000, "loss": 0.5677636433392763, "lr": 0.0003, "grad_norm": 0.14378056862540295, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23563526151, "tokens_per_second": 148497.77133646054, "data_wait_s": 1.337000048806658e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 161100, "loss": 0.5860947147011757, "lr": 0.0003, "grad_norm": 0.1354543352136411, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23577993896, "tokens_per_second": 147977.07669252725, "data_wait_s": 1.991199951589806e-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:db63a498fae2f6fc78e185afeae64c10daa94b976ed6c2ed3d08f3b64760c2f5
-size 128028
+oid sha256:ebb99762502fd8209a20ef2001bdf278277e825fd2b5d7f1853d71eb3b038509
+size 130146