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": 76654.82879960976,
-      "required_pages_per_second_upper_estimate": 0.005303158920064655
+      "required_token_positions_per_second": 75785.70745425079,
+      "required_pages_per_second_upper_estimate": 0.005243031088753362
     },
     {
       "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": 9581.85359995122,
-      "required_pages_per_second_upper_estimate": 0.004011722010443206
+      "required_token_positions_per_second": 9473.213431781349,
+      "required_pages_per_second_upper_estimate": 0.00396623664069517
     },
     {
       "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": 19163.70719990244,
-      "required_pages_per_second_upper_estimate": 0.003991294828795421
+      "required_token_positions_per_second": 18946.426863562698,
+      "required_pages_per_second_upper_estimate": 0.003946041064803652
     },
     {
       "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": 28745.56079985366,
-      "required_pages_per_second_upper_estimate": 0.0017232524780213109
+      "required_token_positions_per_second": 28419.64029534405,
+      "required_pages_per_second_upper_estimate": 0.001703714041427755
     },
     {
       "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": 28745.56079985366,
-      "required_pages_per_second_upper_estimate": 0.0022348409138290815
+      "required_token_positions_per_second": 28419.64029534405,
+      "required_pages_per_second_upper_estimate": 0.0022095020281762538
     },
     {
       "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": 28745.56079985366,
-      "required_pages_per_second_upper_estimate": 0.3687171893620357
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28419.64029534405,
+      "required_pages_per_second_upper_estimate": 0.3645366310763593
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 704566,
-    "producer_batch_seconds": 0.06987772636717438,
-    "producer_p95_seconds": 0.1613607739991494,
-    "gpu_update_seconds": 0.9618180766242261,
-    "producer_batches_per_second": 14.31071175306239,
-    "gpu_batches_per_second": 1.0396976562447071,
-    "producer_headroom": 13.764301253454175,
+    "producer_batch_seconds": 0.306282576148476,
+    "producer_p95_seconds": 0.16079873499984387,
+    "gpu_update_seconds": 0.9728483440562594,
+    "producer_batches_per_second": 3.264958825196873,
+    "gpu_batches_per_second": 1.0279094435526637,
+    "producer_headroom": 3.1763097865046483,
     "sustainable": true,
-    "buffer_seconds": 60.59453882732625,
+    "buffer_seconds": 60.31659733148808,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 398976.3387212232,
-    "upload_bytes_per_second": 3581995.9991216585,
+    "generation_bytes_per_second": 375310.9131167353,
+    "upload_bytes_per_second": 3626857.940900979,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.61103459575196,
+    "estimated_full_shards_per_hour": 80.5329851639418,
     "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": [
@@ -346,6 +346,12 @@
       "bytes": 2065038,
       "pages": 19,
       "shards": 2
+    },
+    {
+      "seconds": 0.9610936170001878,
+      "bytes": 5798019,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -50,3 +50,4 @@
 {"step": 530000, "loss": 0.576223611831665, "lr": 0.0003, "grad_norm": 0.15420024417061787, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77574463850, "tokens_per_second": 152218.1309228728, "data_wait_s": 1.5421000171045307e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 530100, "loss": 0.5909310579299927, "lr": 0.0003, "grad_norm": 0.1518725930758285, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77589143386, "tokens_per_second": 139016.61498685495, "data_wait_s": 1.4861999261484016e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 530200, "loss": 0.5672804117202759, "lr": 0.0003, "grad_norm": 0.1700969856792569, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77603855809, "tokens_per_second": 152030.23211751282, "data_wait_s": 1.2554999557323754e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 530300, "loss": 0.5673822164535522, "lr": 0.0003, "grad_norm": 0.16138348372328465, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77618594937, "tokens_per_second": 153978.77197720777, "data_wait_s": 1.1443999937910121e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:327d5af7a82b3ba98a5a26df6646df3fde0fe588f40b7ad6c2a41f8d7b01abf7
-size 111084
+oid sha256:c6e8e5d989448863d6701d2a330492020b8f365a673d7005118040746ea2fd53
+size 113202