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": 76777.230837837,
-      "required_pages_per_second_upper_estimate": 0.005311626977081074
+      "required_token_positions_per_second": 75420.1236053891,
+      "required_pages_per_second_upper_estimate": 0.005217739149817712
     },
     {
       "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": 9597.153854729626,
-      "required_pages_per_second_upper_estimate": 0.004018127907613272
+      "required_token_positions_per_second": 9427.515450673638,
+      "required_pages_per_second_upper_estimate": 0.003947103850287732
     },
     {
       "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": 19194.30770945925,
-      "required_pages_per_second_upper_estimate": 0.0039976681079463994
+      "required_token_positions_per_second": 18855.030901347276,
+      "required_pages_per_second_upper_estimate": 0.003927005696147787
     },
     {
       "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": 28791.461564188878,
-      "required_pages_per_second_upper_estimate": 0.001726004158756798
+      "required_token_positions_per_second": 28282.546352020912,
+      "required_pages_per_second_upper_estimate": 0.0016954954688558638
     },
     {
       "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": 28791.461564188878,
-      "required_pages_per_second_upper_estimate": 0.0022384094963600193
+      "required_token_positions_per_second": 28282.546352020912,
+      "required_pages_per_second_upper_estimate": 0.002198843577095408
     },
     {
       "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": 28791.461564188878,
-      "required_pages_per_second_upper_estimate": 0.3693059550825269
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28282.546352020912,
+      "required_pages_per_second_upper_estimate": 0.36277813717141794
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.435514724289078,
-    "producer_p95_seconds": 0.29342362099851016,
-    "gpu_update_seconds": 0.9602846989327166,
-    "producer_batches_per_second": 2.296133618977801,
-    "gpu_batches_per_second": 1.0413578401399335,
-    "producer_headroom": 2.204941981009387,
+    "producer_batch_seconds": 0.14372450062495545,
+    "producer_p95_seconds": 0.28017138599898317,
+    "gpu_update_seconds": 0.977564030334363,
+    "producer_batches_per_second": 6.957755954285543,
+    "gpu_batches_per_second": 1.022950895255386,
+    "producer_headroom": 6.801651952754287,
     "sustainable": true,
-    "buffer_seconds": 60.497936032761146,
+    "buffer_seconds": 60.608969880730506,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 287496.0053414626,
+    "generation_bytes_per_second": 285415.1110199606,
     "upload_bytes_per_second": 4807721.9520771345,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.68995018179807,
+    "estimated_full_shards_per_hour": 61.24343870114435,
     "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-v1/1791342976705245012-6808dd50/train.jsonl

@@ -265,3 +265,4 @@
 {"step": 869000, "loss": 0.49112239480018616, "lr": 3e-05, "grad_norm": 0.29584658459382734, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127197639142, "tokens_per_second": 153049.0122783494, "data_wait_s": 1.383099879603833e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 869100, "loss": 0.5332478880882263, "lr": 3e-05, "grad_norm": 0.3049947083564535, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127212320735, "tokens_per_second": 151972.2351776996, "data_wait_s": 1.2565997167257592e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 869200, "loss": 0.5154692530632019, "lr": 3e-05, "grad_norm": 0.30594704435434894, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127226994495, "tokens_per_second": 151496.0445390429, "data_wait_s": 1.036300091072917e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 869300, "loss": 0.5246040225028992, "lr": 3e-05, "grad_norm": 0.3082012577395776, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127241499676, "tokens_per_second": 150615.29035726047, "data_wait_s": 9.862997103482485e-06, "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 sha256:18b00568e5043e32d4f032ccf58e7361ecfa13aacea24785c2a6479b915c12d2
-size 569586
+oid sha256:8f2eb962052c7d4053cc4f16733a25201e0bd5e2d90107c95ea08c76dfebd637
+size 571704