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": 76786.22456593973,
-      "required_pages_per_second_upper_estimate": 0.005312249184059544
+      "required_token_positions_per_second": 75952.44852296314,
+      "required_pages_per_second_upper_estimate": 0.005254566622779465
     },
     {
       "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": 9598.278070742466,
-      "required_pages_per_second_upper_estimate": 0.004018598593381548
+      "required_token_positions_per_second": 9494.056065370392,
+      "required_pages_per_second_upper_estimate": 0.0039749630161352205
     },
     {
       "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": 19196.556141484933,
-      "required_pages_per_second_upper_estimate": 0.003998136397042209
+      "required_token_positions_per_second": 18988.112130740785,
+      "required_pages_per_second_upper_estimate": 0.003954723006642426
     },
     {
       "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": 28794.8342122274,
-      "required_pages_per_second_upper_estimate": 0.001726206343856972
+      "required_token_positions_per_second": 28482.16819611118,
+      "required_pages_per_second_upper_estimate": 0.001707462493604168
     },
     {
       "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": 28794.8342122274,
-      "required_pages_per_second_upper_estimate": 0.0022386717049103035
+      "required_token_positions_per_second": 28482.16819611118,
+      "required_pages_per_second_upper_estimate": 0.002214363297429729
     },
     {
       "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": 28794.8342122274,
-      "required_pages_per_second_upper_estimate": 0.3693492157902977
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28482.16819611118,
+      "required_pages_per_second_upper_estimate": 0.365338671850171
     }
   ],
   "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": 0.10890448349977078,
-    "producer_p95_seconds": 0.31960435799555853,
-    "gpu_update_seconds": 0.9601722238171314,
-    "producer_batches_per_second": 9.18235841045153,
-    "gpu_batches_per_second": 1.0414798253843822,
-    "producer_headroom": 8.816645494849185,
+    "producer_batch_seconds": 0.20598060080442337,
+    "producer_p95_seconds": 0.3176107379986206,
+    "gpu_update_seconds": 0.9707126160351419,
+    "producer_batches_per_second": 4.854826115151933,
+    "gpu_batches_per_second": 1.030171014037586,
+    "producer_headroom": 4.712640958634859,
     "sustainable": true,
-    "buffer_seconds": 60.49085010047928,
+    "buffer_seconds": 60.184182194178796,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 310185.95716764074,
-    "upload_bytes_per_second": 4561313.434877255,
+    "generation_bytes_per_second": 343121.4969128226,
+    "upload_bytes_per_second": 4604400.782809922,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.55868564865033,
+    "estimated_full_shards_per_hour": 73.6258857778407,
     "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": 1.1171102930020425,
-      "bytes": 1254155,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.3684268549986882,
       "bytes": 11656046,
@@ -562,6 +556,12 @@
       "bytes": 866020,
       "pages": 27,
       "shards": 1
+    },
+    {
+      "seconds": 1.6687920610056608,
+      "bytes": 7737148,
+      "pages": 36,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -415,3 +415,4 @@
 {"step": 761500, "loss": 0.5589938163757324, "lr": 5e-05, "grad_norm": 0.2523926632359992, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111461313256, "tokens_per_second": 150659.0759661908, "data_wait_s": 1.3568002032116055e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 761600, "loss": 0.5514082312583923, "lr": 5e-05, "grad_norm": 0.253750221363891, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111475755654, "tokens_per_second": 149425.7521523962, "data_wait_s": 1.4593999367207289e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 761700, "loss": 0.504923403263092, "lr": 5e-05, "grad_norm": 0.23508749085410036, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111490271322, "tokens_per_second": 149424.3890953439, "data_wait_s": 9.561001206748188e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 761800, "loss": 0.5890706777572632, "lr": 5e-05, "grad_norm": 0.27169946887613655, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111505032369, "tokens_per_second": 152226.03419675006, "data_wait_s": 1.207699824590236e-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 sha256:0db3154d4fcf53a296b4e00f6f855b2f0c73f4b06512a58813c25b4ed3a38da4
-size 889635
+oid sha256:4d8b1963fde5ee711d7e0e026d71be1b36f1871bda612e1d77e5e0e46c1776bf
+size 891753