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": 78671.99227083639,
-      "required_pages_per_second_upper_estimate": 0.005442710969468209
+      "required_token_positions_per_second": 78232.09487679884,
+      "required_pages_per_second_upper_estimate": 0.005412277847046112
     },
     {
       "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": 9833.999033854549,
-      "required_pages_per_second_upper_estimate": 0.004117290038223121
+      "required_token_positions_per_second": 9779.011859599856,
+      "required_pages_per_second_upper_estimate": 0.0040942680566254525
     },
     {
       "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": 19667.998067709097,
-      "required_pages_per_second_upper_estimate": 0.004096325317515041
+      "required_token_positions_per_second": 19558.02371919971,
+      "required_pages_per_second_upper_estimate": 0.0040734205609391375
     },
     {
       "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": 29501.997101563647,
-      "required_pages_per_second_upper_estimate": 0.001768599679297469
+      "required_token_positions_per_second": 29337.035578799565,
+      "required_pages_per_second_upper_estimate": 0.0017587104878894232
     },
     {
       "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": 29501.997101563647,
-      "required_pages_per_second_upper_estimate": 0.002293650509075372
+      "required_token_positions_per_second": 29337.035578799565,
+      "required_pages_per_second_upper_estimate": 0.0022808254762695206
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 24,
-      "required_token_positions_per_second": 29501.997101563647,
-      "required_pages_per_second_upper_estimate": 0.37841994204234997
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 29337.035578799565,
+      "required_pages_per_second_upper_estimate": 0.3763039927502157
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,21 +155,21 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 65,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 1.643433355500008,
-    "producer_p95_seconds": 3.2689711630000033,
-    "gpu_update_seconds": 0.9371568949999869,
-    "producer_batches_per_second": 0.6084822342526656,
-    "gpu_batches_per_second": 1.067057186833176,
-    "producer_headroom": 0.5702433213148826,
-    "sustainable": false,
-    "buffer_seconds": 60.91519817499915,
-    "buffer_drain_seconds": 141.7434590228479,
+    "producer_batch_seconds": 0.045079408061542384,
+    "producer_p95_seconds": 0.11253134600002568,
+    "gpu_update_seconds": 0.9424265081499867,
+    "producer_batches_per_second": 22.183077440475717,
+    "gpu_batches_per_second": 1.0610906965711648,
+    "producer_headroom": 20.905920212248276,
+    "sustainable": true,
+    "buffer_seconds": 60.31529652159915,
+    "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 6,
-    "max_pending_pages": 1445,
+    "upload_pages_per_commit": 21,
+    "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,
     "upload_sustainable": null,

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -0,0 +1 @@
+{"step": 720100, "loss": 0.5276756882667542, "lr": 5e-05, "grad_norm": 0.19444800064694195, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105402386187, "tokens_per_second": 151641.9104605178, "data_wait_s": 1.1152999945807096e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:90ccd1a0934fddd691a0254fb4840efb4f195c58139fe7ac6e04d1306b3ffd92
-size 165
+oid sha256:5813b48196590944c28616e15a4d09677959cac6db20e2d59ddeeb23b175de06
+size 2283