Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77065.30960834424,
-      "required_pages_per_second_upper_estimate": 0.005331556934859604
+      "required_token_positions_per_second": 76661.77220946347,
+      "required_pages_per_second_upper_estimate": 0.005303639281269265
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9633.16370104303,
-      "required_pages_per_second_upper_estimate": 0.004033204478293593
+      "required_token_positions_per_second": 9582.721526182933,
+      "required_pages_per_second_upper_estimate": 0.004012085393032063
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19266.32740208606,
-      "required_pages_per_second_upper_estimate": 0.004012667910633547
+      "required_token_positions_per_second": 19165.443052365867,
+      "required_pages_per_second_upper_estimate": 0.003991656361086044
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28899.491103129087,
-      "required_pages_per_second_upper_estimate": 0.0017324803646647086
+      "required_token_positions_per_second": 28748.1645785488,
+      "required_pages_per_second_upper_estimate": 0.0017234085705783444
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28899.491103129087,
-      "required_pages_per_second_upper_estimate": 0.0022468083178408963
+      "required_token_positions_per_second": 28748.1645785488,
+      "required_pages_per_second_upper_estimate": 0.0022350433461698163
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28899.491103129087,
-      "required_pages_per_second_upper_estimate": 0.3706916420149701
+      "required_token_positions_per_second": 28748.1645785488,
+      "required_pages_per_second_upper_estimate": 0.36875058783941717
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.20516314799982638,
-    "producer_p95_seconds": 0.3372377149935346,
-    "gpu_update_seconds": 0.9566950470282302,
-    "producer_batches_per_second": 4.874169702254941,
-    "gpu_batches_per_second": 1.0452651585333148,
-    "producer_headroom": 4.6630940125223646,
+    "producer_batch_seconds": 0.120127787780973,
+    "producer_p95_seconds": 0.3199627790017985,
+    "gpu_update_seconds": 0.9617309628396341,
+    "producer_batches_per_second": 8.32446862189191,
+    "gpu_batches_per_second": 1.0397918322681134,
+    "producer_headroom": 8.005899222860428,
     "sustainable": true,
-    "buffer_seconds": 60.2717879627785,
+    "buffer_seconds": 60.58905065889695,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 531592.012205991,
-    "upload_bytes_per_second": 5799792.100889264,
+    "generation_bytes_per_second": 516703.558364045,
+    "upload_bytes_per_second": 5752391.161093922,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 114.06727099070355,
+    "estimated_full_shards_per_hour": 110.8725553816892,
     "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": 1.2203903480003646,
-      "bytes": 12408575,
-      "pages": 47,
-      "shards": 3
-    },
     {
       "seconds": 1.1006403920000594,
       "bytes": 7578962,
@@ -562,6 +556,12 @@
       "bytes": 7698354,
       "pages": 40,
       "shards": 3
+    },
+    {
+      "seconds": 1.4822090849993401,
+      "bytes": 10438246,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -359,3 +359,4 @@
 {"step": 720900, "loss": 0.5309969186782837, "lr": 8e-05, "grad_norm": 0.21036737659257107, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105519523252, "tokens_per_second": 152467.02468842303, "data_wait_s": 1.1761003406718373e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 721000, "loss": 0.5678845643997192, "lr": 8e-05, "grad_norm": 0.21371042788309177, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105534247906, "tokens_per_second": 152453.4650900134, "data_wait_s": 1.1078001989517361e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 721100, "loss": 0.5659447908401489, "lr": 8e-05, "grad_norm": 0.21239020151595914, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105548900762, "tokens_per_second": 152450.41874510577, "data_wait_s": 1.4537996321450919e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 721200, "loss": 0.5369404554367065, "lr": 8e-05, "grad_norm": 0.21757630034294323, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105563713753, "tokens_per_second": 154697.9656602676, "data_wait_s": 1.6921003407333046e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9020d4b33f85e56f0a692978c6d1282b9ad2d06a03b284aa8ec3ebac361602f7
-size 770244
+oid sha256:efc6f1cb7dcbd8c97265cbee27a26f9812c3eaac7e3b77ed001a6f6fcd2ee90e
+size 772362