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": 75542.49903591366,
-      "required_pages_per_second_upper_estimate": 0.006547832355719608
+      "required_token_positions_per_second": 75510.22954565143,
+      "required_pages_per_second_upper_estimate": 0.006545035318089953
     },
     {
       "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": 9442.812379489207,
-      "required_pages_per_second_upper_estimate": 0.003766879040804694
+      "required_token_positions_per_second": 9438.778693206428,
+      "required_pages_per_second_upper_estimate": 0.003765269943037509
     },
     {
       "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": 18885.624758978414,
-      "required_pages_per_second_upper_estimate": 0.004368856912345936
+      "required_token_positions_per_second": 18877.557386412856,
+      "required_pages_per_second_upper_estimate": 0.004366990667683802
     },
     {
       "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": 28328.437138467623,
-      "required_pages_per_second_upper_estimate": 0.0017250531939095762
+      "required_token_positions_per_second": 28316.336079619283,
+      "required_pages_per_second_upper_estimate": 0.0017243163029150614
     },
     {
       "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": 28328.437138467623,
-      "required_pages_per_second_upper_estimate": 0.002180801509052221
+      "required_token_positions_per_second": 28316.336079619283,
+      "required_pages_per_second_upper_estimate": 0.0021798699360442006
     },
     {
       "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": 28328.437138467623,
-      "required_pages_per_second_upper_estimate": 0.3517443800795612
+      "required_token_positions_per_second": 28316.336079619283,
+      "required_pages_per_second_upper_estimate": 0.35159412542830354
     }
   ],
   "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": 0.04823905089062208,
-    "producer_p95_seconds": 0.16542349899964393,
-    "gpu_update_seconds": 0.9759804208350186,
-    "producer_batches_per_second": 20.73009276794053,
-    "gpu_batches_per_second": 1.0246107182605477,
-    "producer_headroom": 20.232164663603573,
+    "producer_batch_seconds": 0.4320947606875123,
+    "producer_p95_seconds": 0.1710592969998288,
+    "gpu_update_seconds": 0.9763975085710217,
+    "producer_batches_per_second": 2.314307163569596,
+    "gpu_batches_per_second": 1.0241730352871559,
+    "producer_headroom": 2.2596837485774217,
     "sustainable": true,
-    "buffer_seconds": 60.51078609177115,
+    "buffer_seconds": 60.536645531403344,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 360518.29499446676,
+    "generation_bytes_per_second": 324658.6980913005,
     "upload_bytes_per_second": 536068.5311161617,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.35883366942885,
+    "estimated_full_shards_per_hour": 69.66419894270193,
     "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": [

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -26,3 +26,4 @@
 {"step": 42600, "loss": 0.8620638959109783, "lr": 0.0003, "grad_norm": 0.19228546236749122, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6233543216, "tokens_per_second": 150894.77250834717, "data_wait_s": 1.3367000065045431e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 42700, "loss": 0.6080329138785601, "lr": 0.0003, "grad_norm": 0.12684101795425848, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6248313748, "tokens_per_second": 150561.04743861072, "data_wait_s": 1.1466000159998657e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 42800, "loss": 0.6608190611004829, "lr": 0.0003, "grad_norm": 0.14224091184354792, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6263105450, "tokens_per_second": 150208.0042480868, "data_wait_s": 1.0153999937756453e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 42900, "loss": 0.6258081868290901, "lr": 0.0003, "grad_norm": 0.13615424919381686, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6277784532, "tokens_per_second": 149550.8692214105, "data_wait_s": 1.079300000128569e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:18928eacd9653872d9eef10f10909d680abb7a34c264531e84d59e215ca8f560
-size 59469
+oid sha256:8038b7bcd84ce18f909527171a8ba4855a2ed20bf0837f5f4f276ed83c74de28
+size 61587