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": 74464.61849568048,
-      "required_pages_per_second_upper_estimate": 0.00645440440235545
+      "required_token_positions_per_second": 74356.07705713506,
+      "required_pages_per_second_upper_estimate": 0.006444996305558093
     },
     {
       "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": 9308.07731196006,
-      "required_pages_per_second_upper_estimate": 0.00371313120789854
+      "required_token_positions_per_second": 9294.509632141882,
+      "required_pages_per_second_upper_estimate": 0.0037077188575641786
     },
     {
       "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": 18616.15462392012,
-      "required_pages_per_second_upper_estimate": 0.004306519739112599
+      "required_token_positions_per_second": 18589.019264283765,
+      "required_pages_per_second_upper_estimate": 0.004300242451226748
     },
     {
       "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": 27924.231935880176,
-      "required_pages_per_second_upper_estimate": 0.0017004392177728022
+      "required_token_positions_per_second": 27883.528896425647,
+      "required_pages_per_second_upper_estimate": 0.0016979606162223651
     },
     {
       "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": 27924.231935880176,
-      "required_pages_per_second_upper_estimate": 0.002149684673645425
+      "required_token_positions_per_second": 27883.528896425647,
+      "required_pages_per_second_upper_estimate": 0.0021465512409949905
     },
     {
       "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": 27924.231935880176,
-      "required_pages_per_second_upper_estimate": 0.3467255042512159
+      "required_token_positions_per_second": 27883.528896425647,
+      "required_pages_per_second_upper_estimate": 0.3462201087255007
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.07181550292975203,
-    "producer_p95_seconds": 0.27915470199877745,
-    "gpu_update_seconds": 0.9901078054173714,
-    "producer_batches_per_second": 13.924570032993751,
-    "gpu_batches_per_second": 1.0099910277734439,
-    "producer_headroom": 13.786825476747937,
+    "producer_batch_seconds": 0.11278597283600789,
+    "producer_p95_seconds": 0.3048101290005434,
+    "gpu_update_seconds": 0.9915531173510882,
+    "producer_batches_per_second": 8.866350795714744,
+    "gpu_batches_per_second": 1.008518840293173,
+    "producer_headroom": 8.791457771019255,
     "sustainable": true,
-    "buffer_seconds": 60.39657613045966,
+    "buffer_seconds": 60.48474015841638,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 293434.0793217568,
-    "upload_bytes_per_second": 5944817.592390511,
+    "generation_bytes_per_second": 323078.5891720918,
+    "upload_bytes_per_second": 5901394.428871749,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.96412262668159,
+    "estimated_full_shards_per_hour": 69.32514435169283,
     "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": 1.094309231000807,
-      "bytes": 5916543,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.0784175260014308,
       "bytes": 7549095,
@@ -562,6 +556,12 @@
       "bytes": 9937212,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 0.7940713860007236,
+      "bytes": 1067006,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -173,3 +173,4 @@
 {"step": 102300, "loss": 0.6063158456236124, "lr": 0.0003, "grad_norm": 0.13921586346468462, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14970035161, "tokens_per_second": 131085.15676704294, "data_wait_s": 8.13599763205275e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 102400, "loss": 0.6437717862427235, "lr": 0.0003, "grad_norm": 0.14305930577445136, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14984733066, "tokens_per_second": 149375.08400927798, "data_wait_s": 1.2040000001434237e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 102500, "loss": 0.5933273378759623, "lr": 0.0003, "grad_norm": 0.13137340867769348, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14999328157, "tokens_per_second": 148161.86022634414, "data_wait_s": 1.2455999240046367e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 102600, "loss": 0.5845480412244797, "lr": 0.0003, "grad_norm": 0.14226792437529956, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15013821521, "tokens_per_second": 146951.14672099773, "data_wait_s": 1.2579002941492945e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5d28bf471cf07de8f5d598d90f6077b584b518b3c6eac1058111e64b3094c226
-size 373164
+oid sha256:a634c43b05465e03b8f3fad6e5e4ad52a632757402bb2addfd844c37730f00a6
+size 375282