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": 75604.60130486506,
-      "required_pages_per_second_upper_estimate": 0.006553215222995555
+      "required_token_positions_per_second": 75465.95916257486,
+      "required_pages_per_second_upper_estimate": 0.006541198073487136
     },
     {
       "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": 9450.575163108133,
-      "required_pages_per_second_upper_estimate": 0.0037699757312542417
+      "required_token_positions_per_second": 9433.244895321857,
+      "required_pages_per_second_upper_estimate": 0.0037630624283237026
     },
     {
       "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": 18901.150326216266,
-      "required_pages_per_second_upper_estimate": 0.004372448479085762
+      "required_token_positions_per_second": 18866.489790643715,
+      "required_pages_per_second_upper_estimate": 0.004364430374185644
     },
     {
       "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": 28351.725489324395,
-      "required_pages_per_second_upper_estimate": 0.0017264713322922196
+      "required_token_positions_per_second": 28299.734685965574,
+      "required_pages_per_second_upper_estimate": 0.0017233053651423314
     },
     {
       "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": 28351.725489324395,
-      "required_pages_per_second_upper_estimate": 0.0021825943107674564
+      "required_token_positions_per_second": 28299.734685965574,
+      "required_pages_per_second_upper_estimate": 0.0021785919148051393
     },
     {
       "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": 28351.725489324395,
-      "required_pages_per_second_upper_estimate": 0.3520335434561058
+      "required_token_positions_per_second": 28299.734685965574,
+      "required_pages_per_second_upper_estimate": 0.3513879916804149
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.09656628517964805,
-    "producer_p95_seconds": 0.3321660140027234,
-    "gpu_update_seconds": 0.9751787421337238,
-    "producer_batches_per_second": 10.355581123780832,
-    "gpu_batches_per_second": 1.0254530341914205,
-    "producer_headroom": 10.098542574352326,
+    "producer_batch_seconds": 0.47689373899214615,
+    "producer_p95_seconds": 0.31537712499994086,
+    "gpu_update_seconds": 0.9769702898914886,
+    "producer_batches_per_second": 2.096903184582318,
+    "gpu_batches_per_second": 1.023572579787528,
+    "producer_headroom": 2.048612112115773,
     "sustainable": true,
-    "buffer_seconds": 60.46108201229087,
+    "buffer_seconds": 60.57215797327229,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 460237.0836897882,
-    "upload_bytes_per_second": 5560398.826621847,
+    "generation_bytes_per_second": 397049.1279904722,
+    "upload_bytes_per_second": 5512503.396050706,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 98.75616438884958,
+    "estimated_full_shards_per_hour": 85.19750003610253,
     "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": 1.4370965530015383,
-      "bytes": 7562097,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 0.8747751649989368,
       "bytes": 1118624,
@@ -562,6 +556,12 @@
       "bytes": 6234779,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.154916157000116,
+      "bytes": 2367312,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -216,3 +216,4 @@
 {"step": 258700, "loss": 0.5411769356578588, "lr": 0.0003, "grad_norm": 0.1307084377369494, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37866358095, "tokens_per_second": 146500.80926690865, "data_wait_s": 7.2500006353948265e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 258800, "loss": 0.5922668986022472, "lr": 0.0003, "grad_norm": 0.1422421201777596, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37880753239, "tokens_per_second": 147131.9444011959, "data_wait_s": 1.2044001778122038e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 258900, "loss": 0.5958917699754238, "lr": 0.0003, "grad_norm": 0.15891095911456893, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37895376980, "tokens_per_second": 148065.38963443547, "data_wait_s": 1.373000122839585e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 259000, "loss": 0.5608449708670378, "lr": 0.0003, "grad_norm": 0.15565372289390136, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37910074962, "tokens_per_second": 127898.83528886373, "data_wait_s": 1.0024999937741086e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4ecf111d2a2f9445091b41100c2927be872ec81c3975794ff4f3b27f80cba486
-size 465021
+oid sha256:efb82bca3749616183ad5473250b792ba0481b3006ef7a522603550ec78b1ef4
+size 467139