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": 77158.32437170018,
-      "required_pages_per_second_upper_estimate": 0.005337991912012553
+      "required_token_positions_per_second": 76855.45733466595,
+      "required_pages_per_second_upper_estimate": 0.00531703886764218
     },
     {
       "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": 9644.790546462522,
-      "required_pages_per_second_upper_estimate": 0.004038072395674593
+      "required_token_positions_per_second": 9606.932166833243,
+      "required_pages_per_second_upper_estimate": 0.00402222188269668
     },
     {
       "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": 19289.581092925044,
-      "required_pages_per_second_upper_estimate": 0.004017511041194242
+      "required_token_positions_per_second": 19213.864333666486,
+      "required_pages_per_second_upper_estimate": 0.004001741237025904
     },
     {
       "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": 28934.37163938757,
-      "required_pages_per_second_upper_estimate": 0.0017345713995539096
+      "required_token_positions_per_second": 28820.796500499728,
+      "required_pages_per_second_upper_estimate": 0.0017277627433967793
     },
     {
       "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": 28934.37163938757,
-      "required_pages_per_second_upper_estimate": 0.0022495201261117288
+      "required_token_positions_per_second": 28820.796500499728,
+      "required_pages_per_second_upper_estimate": 0.0022406901586274394
     },
     {
       "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": 28934.37163938757,
-      "required_pages_per_second_upper_estimate": 0.37113905208229203
+      "required_token_positions_per_second": 28820.796500499728,
+      "required_pages_per_second_upper_estimate": 0.36968223214812185
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.0665144865781997,
-    "producer_p95_seconds": 0.1507300370003577,
-    "gpu_update_seconds": 0.9555417461481532,
-    "producer_batches_per_second": 15.034318859611446,
-    "gpu_batches_per_second": 1.046526752003312,
-    "producer_headroom": 14.365919295261232,
+    "producer_batch_seconds": 0.25262821601570806,
+    "producer_p95_seconds": 0.1649897269999201,
+    "gpu_update_seconds": 0.9593072835277334,
+    "producer_batches_per_second": 3.958386025802523,
+    "gpu_batches_per_second": 1.042418854908121,
+    "producer_headroom": 3.797308545566759,
     "sustainable": true,
-    "buffer_seconds": 60.19913000733365,
+    "buffer_seconds": 60.436358862247204,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 482544.742597661,
+    "generation_bytes_per_second": 456349.31770231854,
     "upload_bytes_per_second": 2564674.01380149,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 103.54286869475719,
+    "estimated_full_shards_per_hour": 97.92194031049888,
     "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": [

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

@@ -32,3 +32,4 @@
 {"step": 528200, "loss": 0.5649438500404358, "lr": 0.0003, "grad_norm": 0.15943699597456476, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77310758756, "tokens_per_second": 151397.26698401864, "data_wait_s": 1.5206000171019696e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 528300, "loss": 0.5714080333709717, "lr": 0.0003, "grad_norm": 0.16946352881597554, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77325439916, "tokens_per_second": 153162.9703753375, "data_wait_s": 1.6467000023112632e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 528400, "loss": 0.5965216755867004, "lr": 0.0003, "grad_norm": 0.14731440593263612, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77340061987, "tokens_per_second": 151818.14623003325, "data_wait_s": 1.2556000001495704e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 528500, "loss": 0.644488513469696, "lr": 0.0003, "grad_norm": 0.17564630383103438, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77354750227, "tokens_per_second": 153199.0555196711, "data_wait_s": 1.1514000107126776e-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 sha256:7e6965c1d92a2d73758574a8cfa54a236dca0c1f1efa65584dc01e9b5c470b9e
-size 72177
+oid sha256:90429f936f41b0c5a118be73fb65e01867c1b3ca94e5a7fa4cdfbe0a68d731ac
+size 74295