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": 74212.5556820602,
-      "required_pages_per_second_upper_estimate": 0.006432556236518205
+      "required_token_positions_per_second": 73925.70861780913,
+      "required_pages_per_second_upper_estimate": 0.006407693059996159
     },
     {
       "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": 9276.569460257526,
-      "required_pages_per_second_upper_estimate": 0.003700562254770036
+      "required_token_positions_per_second": 9240.71357722614,
+      "required_pages_per_second_upper_estimate": 0.0036862588069355917
     },
     {
       "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": 18553.13892051505,
-      "required_pages_per_second_upper_estimate": 0.004291942165168339
+      "required_token_positions_per_second": 18481.42715445228,
+      "required_pages_per_second_upper_estimate": 0.004275352910173694
     },
     {
       "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": 27829.70838077258,
-      "required_pages_per_second_upper_estimate": 0.001694683229193516
+      "required_token_positions_per_second": 27722.14073167842,
+      "required_pages_per_second_upper_estimate": 0.0016881329237275202
     },
     {
       "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": 27829.70838077258,
-      "required_pages_per_second_upper_estimate": 0.002142407988715295
+      "required_token_positions_per_second": 27722.14073167842,
+      "required_pages_per_second_upper_estimate": 0.0021341271333217156
     },
     {
       "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": 27829.70838077258,
-      "required_pages_per_second_upper_estimate": 0.3455518380467683
+      "required_token_positions_per_second": 27722.14073167842,
+      "required_pages_per_second_upper_estimate": 0.3442162078507819
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.12812085752339897,
-    "producer_p95_seconds": 0.17923944799986202,
-    "gpu_update_seconds": 0.9934707048206757,
-    "producer_batches_per_second": 7.80513040054675,
-    "gpu_batches_per_second": 1.0065722070591934,
-    "producer_headroom": 7.754168400248462,
+    "producer_batch_seconds": 0.09400759942963788,
+    "producer_p95_seconds": 0.1841930250002406,
+    "gpu_update_seconds": 0.9973255769676654,
+    "producer_batches_per_second": 10.637437888715292,
+    "gpu_batches_per_second": 1.0026815947511003,
+    "producer_headroom": 10.608988879820684,
     "sustainable": true,
-    "buffer_seconds": 60.60171299406122,
+    "buffer_seconds": 60.83686019502759,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 383863.90580676746,
-    "upload_bytes_per_second": 5223763.296938822,
+    "generation_bytes_per_second": 384088.068198128,
+    "upload_bytes_per_second": 5297684.240310406,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.36825829174298,
+    "estimated_full_shards_per_hour": 82.41635832269554,
     "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": [
@@ -400,6 +400,12 @@
       "bytes": 6105703,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.3559997279999152,
+      "bytes": 10442134,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -44,3 +44,4 @@
 {"step": 319400, "loss": 0.5801581870764494, "lr": 0.0003, "grad_norm": 0.1494298392152412, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46751229030, "tokens_per_second": 146928.5534759068, "data_wait_s": 1.0874999134102836e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 319500, "loss": 0.5966622941195965, "lr": 0.0003, "grad_norm": 0.15670214870774235, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46765904094, "tokens_per_second": 148631.34836431436, "data_wait_s": 1.509100002294872e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 319600, "loss": 0.5572373159229755, "lr": 0.0003, "grad_norm": 0.16812903759002704, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46780614063, "tokens_per_second": 148200.54024997004, "data_wait_s": 1.2200000128359534e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 319700, "loss": 0.6266476754099131, "lr": 0.0003, "grad_norm": 0.1504831847851161, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46795214701, "tokens_per_second": 148718.9627040397, "data_wait_s": 1.4939000720914919e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a9bd6e81c7b20a6b0de88bd0dc73b27fea387595261184346bde2db7d904cf53
-size 97593
+oid sha256:2b02cec09e6debdda3282557585351d9bbc5ebc2d11c86674f7f6be280201ec4
+size 99711