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": 74839.40138715871,
-      "required_pages_per_second_upper_estimate": 0.006486889633510229
+      "required_token_positions_per_second": 74357.15888876183,
+      "required_pages_per_second_upper_estimate": 0.006445090075981624
     },
     {
       "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": 9354.925173394839,
-      "required_pages_per_second_upper_estimate": 0.0037318195202628206
+      "required_token_positions_per_second": 9294.644861095228,
+      "required_pages_per_second_upper_estimate": 0.003707772802415521
     },
     {
       "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": 18709.850346789677,
-      "required_pages_per_second_upper_estimate": 0.0043281945956100696
+      "required_token_positions_per_second": 18589.289722190457,
+      "required_pages_per_second_upper_estimate": 0.004300305016903564
     },
     {
       "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": 28064.775520184514,
-      "required_pages_per_second_upper_estimate": 0.001708997584681737
+      "required_token_positions_per_second": 27883.934583285685,
+      "required_pages_per_second_upper_estimate": 0.0016979853204236671
     },
     {
       "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": 28064.775520184514,
-      "required_pages_per_second_upper_estimate": 0.0021605041078147143
+      "required_token_positions_per_second": 27883.934583285685,
+      "required_pages_per_second_upper_estimate": 0.0021465824718924892
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 28064.775520184514,
-      "required_pages_per_second_upper_estimate": 0.3484705851991571
+      "required_token_positions_per_second": 27883.934583285685,
+      "required_pages_per_second_upper_estimate": 0.34622514599855575
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.10224816711722085,
-    "producer_p95_seconds": 0.14535556900000302,
-    "gpu_update_seconds": 0.9851495152745916,
-    "producer_batches_per_second": 9.78012641393919,
-    "gpu_batches_per_second": 1.0150743460714886,
-    "producer_headroom": 9.634886796016422,
+    "producer_batch_seconds": 0.07211896839061893,
+    "producer_p95_seconds": 0.15178614300020854,
+    "gpu_update_seconds": 0.9915386911204738,
+    "producer_batches_per_second": 13.865977596680066,
+    "gpu_batches_per_second": 1.008533513573701,
+    "producer_headroom": 13.748653277317967,
     "sustainable": true,
-    "buffer_seconds": 60.094120431750085,
+    "buffer_seconds": 60.4838601583489,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 536636.6000627237,
-    "upload_bytes_per_second": 5374126.192712446,
+    "generation_bytes_per_second": 535769.0341422006,
+    "upload_bytes_per_second": 5350153.571532421,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 115.14972211276326,
+    "estimated_full_shards_per_hour": 114.96356266212,
     "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": [
@@ -280,6 +280,12 @@
       "bytes": 2243299,
       "pages": 26,
       "shards": 2
+    },
+    {
+      "seconds": 1.2289755599999808,
+      "bytes": 6161182,
+      "pages": 34,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790476562033999719-ba4f58e1/train.jsonl

@@ -17,3 +17,4 @@
 {"step": 83600, "loss": 0.6076688952744007, "lr": 0.0003, "grad_norm": 0.13999636592620193, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12234408915, "tokens_per_second": 149894.32157979623, "data_wait_s": 1.3642999874718953e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 83700, "loss": 0.5898118000477552, "lr": 0.0003, "grad_norm": 0.13819632783180416, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12248905316, "tokens_per_second": 146654.57550567205, "data_wait_s": 1.4201000340108294e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 83800, "loss": 0.5947676505893469, "lr": 0.0003, "grad_norm": 0.1317681750560417, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12263476388, "tokens_per_second": 147484.64133959956, "data_wait_s": 1.4195999938237946e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 83900, "loss": 0.6639142297208309, "lr": 0.0003, "grad_norm": 0.14489169903113333, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12278239593, "tokens_per_second": 149528.6117216716, "data_wait_s": 1.132099987444235e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790476562.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.417.0.1790476562033999719-ba4f58e1

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:902758550c089305c5328e444efc8d5180913a350074c2edb5a1edf5df860d60
-size 40407
+oid sha256:cf846ca144802eccaa00a055f167b59d70f4960c69b7ad4cc6cc31cd0b718e60
+size 42525