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": 76985.660131269,
-      "required_pages_per_second_upper_estimate": 0.00532604659922329
+      "required_token_positions_per_second": 75731.21529867845,
+      "required_pages_per_second_upper_estimate": 0.005239261194991646
     },
     {
       "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": 9623.207516408625,
-      "required_pages_per_second_upper_estimate": 0.004029036031695946
+      "required_token_positions_per_second": 9466.401912334806,
+      "required_pages_per_second_upper_estimate": 0.003963384799743603
     },
     {
       "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": 19246.41503281725,
-      "required_pages_per_second_upper_estimate": 0.004008520689239345
+      "required_token_positions_per_second": 18932.803824669612,
+      "required_pages_per_second_upper_estimate": 0.003943203745065917
     },
     {
       "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": 28869.622549225878,
-      "required_pages_per_second_upper_estimate": 0.0017306897904648523
+      "required_token_positions_per_second": 28399.20573700442,
+      "required_pages_per_second_upper_estimate": 0.0017024890208570582
     },
     {
       "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": 28869.622549225878,
-      "required_pages_per_second_upper_estimate": 0.002244486169152806
+      "required_token_positions_per_second": 28399.20573700442,
+      "required_pages_per_second_upper_estimate": 0.0022079133311474707
     },
     {
       "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": 23,
-      "required_token_positions_per_second": 28869.622549225878,
-      "required_pages_per_second_upper_estimate": 0.3703085202758543
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28399.20573700442,
+      "required_pages_per_second_upper_estimate": 0.3642745185028979
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.11384489394498587,
-    "producer_p95_seconds": 0.3285287629987579,
-    "gpu_update_seconds": 0.9576848451294132,
-    "producer_batches_per_second": 8.783880992354717,
-    "gpu_batches_per_second": 1.044184843360311,
-    "producer_headroom": 8.412189707798424,
+    "producer_batch_seconds": 0.1320987807341112,
+    "producer_p95_seconds": 0.3112501530049485,
+    "gpu_update_seconds": 0.973548353994084,
+    "producer_batches_per_second": 7.570092581042084,
+    "gpu_batches_per_second": 1.0271703463904955,
+    "producer_headroom": 7.3698511718563475,
     "sustainable": true,
-    "buffer_seconds": 60.33414524315303,
+    "buffer_seconds": 60.35999794763321,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 393883.1497213826,
-    "upload_bytes_per_second": 4113133.833290097,
+    "generation_bytes_per_second": 390666.22237106087,
+    "upload_bytes_per_second": 4111056.838902051,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.51815479975804,
+    "estimated_full_shards_per_hour": 83.82787707661504,
     "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": 1.9317821920012648,
-      "bytes": 6082541,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.9298742110004241,
       "bytes": 11871053,
@@ -562,6 +556,12 @@
       "bytes": 6111417,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 1.0848596139985602,
+      "bytes": 2384316,
+      "pages": 30,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -332,3 +332,4 @@
 {"step": 875700, "loss": 0.5712859034538269, "lr": 3e-05, "grad_norm": 0.33275925961525316, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128177914717, "tokens_per_second": 154362.46005396164, "data_wait_s": 2.579300053184852e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 875800, "loss": 0.5627284646034241, "lr": 3e-05, "grad_norm": 0.326746899017602, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128192621469, "tokens_per_second": 152486.04017003742, "data_wait_s": 1.3174001651350409e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 875900, "loss": 0.5297982692718506, "lr": 3e-05, "grad_norm": 0.31422191003205474, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128207114873, "tokens_per_second": 151192.09840029027, "data_wait_s": 1.1267999070696533e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 876000, "loss": 0.532098650932312, "lr": 3e-05, "grad_norm": 0.2955324069768495, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128221751007, "tokens_per_second": 151555.89544821964, "data_wait_s": 2.1441999706439674e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8c0269c7cfffd610fd9c4febc60c8ce104a134e901dd490d199ac4fb5685731a
-size 713058
+oid sha256:c2f5cacd1212ed991bdcabd5e405508816957cdf17b16518addb9426aef32bb1
+size 715176