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": 76070.78323781662,
-      "required_pages_per_second_upper_estimate": 0.005262753292927396
+      "required_token_positions_per_second": 76803.17404064437,
+      "required_pages_per_second_upper_estimate": 0.0053134217880478115
     },
     {
       "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": 9508.847904727078,
-      "required_pages_per_second_upper_estimate": 0.003981156050385133
+      "required_token_positions_per_second": 9600.396755080546,
+      "required_pages_per_second_upper_estimate": 0.004019485642270742
     },
     {
       "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": 19017.695809454155,
-      "required_pages_per_second_upper_estimate": 0.003960884506744349
+      "required_token_positions_per_second": 19200.793510161093,
+      "required_pages_per_second_upper_estimate": 0.003999018929190527
     },
     {
       "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": 28526.54371418123,
-      "required_pages_per_second_upper_estimate": 0.0017101227381549748
+      "required_token_positions_per_second": 28801.19026524164,
+      "required_pages_per_second_upper_estimate": 0.0017265873795300447
     },
     {
       "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": 28526.54371418123,
-      "required_pages_per_second_upper_estimate": 0.002217813298772399
+      "required_token_positions_per_second": 28801.19026524164,
+      "required_pages_per_second_upper_estimate": 0.0022391658600748377
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28526.54371418123,
-      "required_pages_per_second_upper_estimate": 0.3659078733492545
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28801.19026524164,
+      "required_pages_per_second_upper_estimate": 0.3694307444137664
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.2554650053047567,
-    "producer_p95_seconds": 0.3141337189990736,
-    "gpu_update_seconds": 0.9692025881935187,
-    "producer_batches_per_second": 3.9144304669324517,
-    "gpu_batches_per_second": 1.0317760313288928,
-    "producer_headroom": 3.793876139854496,
+    "producer_batch_seconds": 0.08154234182038067,
+    "producer_p95_seconds": 0.2968129239998234,
+    "gpu_update_seconds": 0.959960326131613,
+    "producer_batches_per_second": 12.263567340300009,
+    "gpu_batches_per_second": 1.0417097173481495,
+    "producer_headroom": 11.772538103531394,
     "sustainable": true,
-    "buffer_seconds": 60.09056046799816,
+    "buffer_seconds": 60.477500546291616,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 358997.4777137333,
-    "upload_bytes_per_second": 4710007.531903036,
+    "generation_bytes_per_second": 345666.0653358303,
+    "upload_bytes_per_second": 4841759.838743776,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.03250168379782,
+    "estimated_full_shards_per_hour": 74.17189092689688,
     "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": 2.7016539569995075,
-      "bytes": 7555720,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 1.483174205000978,
       "bytes": 5900678,
@@ -562,6 +556,12 @@
       "bytes": 5015589,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.4427611729988712,
+      "bytes": 13297372,
+      "pages": 36,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -281,3 +281,4 @@
 {"step": 870600, "loss": 0.5031870603561401, "lr": 3e-05, "grad_norm": 0.30149928247060004, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127431503243, "tokens_per_second": 149693.02054720913, "data_wait_s": 1.001199780148454e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 870700, "loss": 0.520969808101654, "lr": 3e-05, "grad_norm": 0.30597625598022377, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127446027794, "tokens_per_second": 150416.08419461123, "data_wait_s": 1.535900082672015e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 870800, "loss": 0.49362891912460327, "lr": 3e-05, "grad_norm": 0.30498236498352627, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127460642191, "tokens_per_second": 151942.53239090004, "data_wait_s": 1.5082998288562521e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 870900, "loss": 0.5143728256225586, "lr": 3e-05, "grad_norm": 0.29668275921008563, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127475238859, "tokens_per_second": 151386.29090419709, "data_wait_s": 1.0212999768555164e-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 sha256:4c285b2de53581e2b9f4c4c825cbb4e2ce9d88dc2a6e865c6e88d56fab23ac3b
-size 604257
+oid sha256:263621ed126a3623d1878b7668ab5ef1ea7955996740eb6d2cba5b2cb5022c28
+size 606375