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": 76017.94633184573,
-      "required_pages_per_second_upper_estimate": 0.006589042922320845
+      "required_token_positions_per_second": 75917.17339470465,
+      "required_pages_per_second_upper_estimate": 0.006580308179536131
     },
     {
       "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": 9502.243291480716,
-      "required_pages_per_second_upper_estimate": 0.0037905869201694253
+      "required_token_positions_per_second": 9489.64667433808,
+      "required_pages_per_second_upper_estimate": 0.0037855619412550186
     },
     {
       "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": 19004.486582961432,
-      "required_pages_per_second_upper_estimate": 0.004396353503427752
+      "required_token_positions_per_second": 18979.29334867616,
+      "required_pages_per_second_upper_estimate": 0.004390525492061635
     },
     {
       "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": 28506.72987444215,
-      "required_pages_per_second_upper_estimate": 0.001735910286100733
+      "required_token_positions_per_second": 28468.940023014242,
+      "required_pages_per_second_upper_estimate": 0.001733609082416801
     },
     {
       "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": 28506.72987444215,
-      "required_pages_per_second_upper_estimate": 0.0021945269774134215
+      "required_token_positions_per_second": 28468.940023014242,
+      "required_pages_per_second_upper_estimate": 0.002191617810041497
     },
     {
       "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": 28506.72987444215,
-      "required_pages_per_second_upper_estimate": 0.3539581791529626
+      "required_token_positions_per_second": 28468.940023014242,
+      "required_pages_per_second_upper_estimate": 0.35348895567272487
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.15307781606213666,
-    "producer_p95_seconds": 0.39527264700154774,
-    "gpu_update_seconds": 0.9698762405149793,
-    "producer_batches_per_second": 6.532625208045067,
-    "gpu_batches_per_second": 1.031059384926293,
-    "producer_headroom": 6.335837977472134,
+    "producer_batch_seconds": 0.10167211079664185,
+    "producer_p95_seconds": 0.38399030399887124,
+    "gpu_update_seconds": 0.9711636603838922,
+    "producer_batches_per_second": 9.835538892274373,
+    "gpu_batches_per_second": 1.0296925644898092,
+    "producer_headroom": 9.551917952469312,
     "sustainable": true,
-    "buffer_seconds": 60.13232691192872,
+    "buffer_seconds": 60.212146943801315,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 298670.4634382475,
+    "generation_bytes_per_second": 298328.0392779603,
     "upload_bytes_per_second": 5292592.716316272,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.08772876129693,
+    "estimated_full_shards_per_hour": 64.01425250772579,
     "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-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -376,3 +376,4 @@
 {"step": 274700, "loss": 0.5999009869992733, "lr": 0.0003, "grad_norm": 0.168274931024671, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40207522222, "tokens_per_second": 150018.71773799387, "data_wait_s": 9.433999366592616e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 274800, "loss": 0.6245676893740892, "lr": 0.0003, "grad_norm": 0.14362337745051237, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40222223676, "tokens_per_second": 125160.99911783844, "data_wait_s": 9.800998668652028e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 274900, "loss": 0.6254910230636597, "lr": 0.0003, "grad_norm": 0.16177351706420479, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40236919200, "tokens_per_second": 150220.9985099321, "data_wait_s": 1.0051997378468513e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 275000, "loss": 0.5781365633010864, "lr": 0.0003, "grad_norm": 0.16510326409711035, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40251539766, "tokens_per_second": 148894.20088002158, "data_wait_s": 1.0673000360839069e-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:e3c9e50d2f635cf076668864a14b0104d34f47937bc91530c39f765bd9bfdf70
-size 806250
+oid sha256:685d89dc729b9c384c57b9c9578291e84c459bda5f68cbf575f5f6ea27cf2b59
+size 808368