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": 76626.63431968096,
-      "required_pages_per_second_upper_estimate": 0.005301208360522974
+      "required_token_positions_per_second": 76388.93169665427,
+      "required_pages_per_second_upper_estimate": 0.005284763541515914
     },
     {
       "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": 9578.32928996012,
-      "required_pages_per_second_upper_estimate": 0.004010246455445893
+      "required_token_positions_per_second": 9548.616462081784,
+      "required_pages_per_second_upper_estimate": 0.003997806314887636
     },
     {
       "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": 19156.65857992024,
-      "required_pages_per_second_upper_estimate": 0.003989826787137737
+      "required_token_positions_per_second": 19097.232924163567,
+      "required_pages_per_second_upper_estimate": 0.003977449990203551
     },
     {
       "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": 28734.987869880362,
-      "required_pages_per_second_upper_estimate": 0.0017226186470133409
+      "required_token_positions_per_second": 28645.84938624535,
+      "required_pages_per_second_upper_estimate": 0.0017172749310260056
     },
     {
       "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": 28734.987869880362,
-      "required_pages_per_second_upper_estimate": 0.0022340189150290616
+      "required_token_positions_per_second": 28645.84938624535,
+      "required_pages_per_second_upper_estimate": 0.0022270887900051927
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28734.987869880362,
-      "required_pages_per_second_upper_estimate": 0.3685815711686659
+      "required_token_positions_per_second": 28645.84938624535,
+      "required_pages_per_second_upper_estimate": 0.3674381984100428
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.22836590607806784,
+    "producer_batch_seconds": 0.12022614786718577,
     "producer_p95_seconds": 0.33813577800174244,
-    "gpu_update_seconds": 0.962171973943315,
-    "producer_batches_per_second": 4.378937369303042,
-    "gpu_batches_per_second": 1.0393152441362978,
-    "producer_headroom": 4.213290812396455,
+    "gpu_update_seconds": 0.9651660045826925,
+    "producer_batches_per_second": 8.317658161223823,
+    "gpu_batches_per_second": 1.0360911959724157,
+    "producer_headroom": 8.027920894953024,
     "sustainable": true,
-    "buffer_seconds": 60.61683435842885,
+    "buffer_seconds": 60.80545828870963,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "shard_target_mib": 32,
+    "upload_pages_per_commit": 167,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 509262.29149063875,
-    "upload_bytes_per_second": 4566447.164578904,
+    "generation_bytes_per_second": 656497.609361106,
+    "upload_bytes_per_second": 4514500.470039032,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 109.27583273448344,
+    "estimated_full_shards_per_hour": 70.43455224335138,
     "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": 1.0989994560004561,
-      "bytes": 10674348,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.3233692020003218,
       "bytes": 11880619,
@@ -562,6 +556,12 @@
       "bytes": 6493791,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.117266364999523,
+      "bytes": 5833475,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -427,3 +427,4 @@
 {"step": 762700, "loss": 0.5550649762153625, "lr": 5e-05, "grad_norm": 0.27113370539924303, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111636742854, "tokens_per_second": 152260.83182660094, "data_wait_s": 6.713999027851969e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 762800, "loss": 0.5426653623580933, "lr": 5e-05, "grad_norm": 0.23226449571497457, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111651462115, "tokens_per_second": 151931.079185711, "data_wait_s": 1.3764998584520072e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 762900, "loss": 0.5179505348205566, "lr": 5e-05, "grad_norm": 0.2444206139901824, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111666008715, "tokens_per_second": 149836.23805001707, "data_wait_s": 1.2472999515011907e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 763000, "loss": 0.5827005505561829, "lr": 5e-05, "grad_norm": 0.23648758763056318, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111680576336, "tokens_per_second": 150520.9669195791, "data_wait_s": 1.6065998352132738e-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 sha256:72644c51798a0cfe5f8f6d8e2d6d4b774d9e430617b341b39d8f9233c41e7827
-size 915051
+oid sha256:a6c9d2786a6e3a90f98712ba52735ffa0171311a4f051514df416e8780c4308c
+size 917169