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": 77141.16048393378,
-      "required_pages_per_second_upper_estimate": 0.005336804474431168
+      "required_token_positions_per_second": 77007.70364335265,
+      "required_pages_per_second_upper_estimate": 0.005327571620537232
     },
     {
       "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": 9642.645060491723,
-      "required_pages_per_second_upper_estimate": 0.004037174125501462
+      "required_token_positions_per_second": 9625.962955419081,
+      "required_pages_per_second_upper_estimate": 0.004030189676469514
     },
     {
       "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": 19285.290120983445,
-      "required_pages_per_second_upper_estimate": 0.004016617344899347
+      "required_token_positions_per_second": 19251.925910838163,
+      "required_pages_per_second_upper_estimate": 0.004009668459799475
     },
     {
       "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": 28927.935181475164,
-      "required_pages_per_second_upper_estimate": 0.001734185543729963
+      "required_token_positions_per_second": 28877.888866257243,
+      "required_pages_per_second_upper_estimate": 0.0017311853435489474
     },
     {
       "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": 28927.935181475164,
-      "required_pages_per_second_upper_estimate": 0.0022490197198200253
+      "required_token_positions_per_second": 28877.888866257243,
+      "required_pages_per_second_upper_estimate": 0.0022451288389423087
     },
     {
       "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": 28927.935181475164,
-      "required_pages_per_second_upper_estimate": 0.3710564921111218
+      "required_token_positions_per_second": 28877.888866257243,
+      "required_pages_per_second_upper_estimate": 0.37041455171505294
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.08364896867969662,
-    "producer_p95_seconds": 0.17260160100067878,
-    "gpu_update_seconds": 0.9557543539334667,
-    "producer_batches_per_second": 11.95472001369362,
-    "gpu_batches_per_second": 1.046293951876272,
-    "producer_headroom": 11.42577570314323,
+    "producer_batch_seconds": 0.06769080817188211,
+    "producer_p95_seconds": 0.16766758199992182,
+    "gpu_update_seconds": 0.9574107071346782,
+    "producer_batches_per_second": 14.773054525524,
+    "gpu_batches_per_second": 1.0444838276279385,
+    "producer_headroom": 14.143880579821092,
     "sustainable": true,
-    "buffer_seconds": 60.21252429780841,
+    "buffer_seconds": 60.31687454948473,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 32,
-    "upload_pages_per_commit": 155,
+    "shard_target_mib": 16,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 565117.0559436153,
+    "generation_bytes_per_second": 557284.1809064293,
     "upload_bytes_per_second": 5446512.698423026,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.63048247686073,
+    "estimated_full_shards_per_hour": 119.58021231074008,
     "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-v1/1791039222312944374-63c10d34/train.jsonl

@@ -44,3 +44,4 @@
 {"step": 608800, "loss": 0.5435750484466553, "lr": 0.0001, "grad_norm": 0.15740634315374563, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89110419345, "tokens_per_second": 153860.36729184573, "data_wait_s": 1.3574000149674248e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 608900, "loss": 0.5549183487892151, "lr": 0.0001, "grad_norm": 0.136732323850195, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89125119530, "tokens_per_second": 152651.59915366335, "data_wait_s": 1.2350000361038838e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 609000, "loss": 0.5282745957374573, "lr": 0.0001, "grad_norm": 0.13622988014272464, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89139926638, "tokens_per_second": 153269.223486098, "data_wait_s": 1.1479000022518449e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 609100, "loss": 0.5480009913444519, "lr": 0.0001, "grad_norm": 0.13478104810810157, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89154716094, "tokens_per_second": 153392.15708586172, "data_wait_s": 1.6027000128815416e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6793f3ef224ba655820340b0e6af515e0730934b5b7a5d56d35bd933695b9d9d
-size 98376
+oid sha256:ff315ec42e09300410ac4fd42704eb0e49b633f6fc97d2a84e6d54a6c9707352
+size 100494