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": 77154.66575258688,
-      "required_pages_per_second_upper_estimate": 0.00533773880025313
+      "required_token_positions_per_second": 77343.8109768526,
+      "required_pages_per_second_upper_estimate": 0.005350824305745219
     },
     {
       "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": 9644.33321907336,
-      "required_pages_per_second_upper_estimate": 0.004037880922246833
+      "required_token_positions_per_second": 9667.976372106576,
+      "required_pages_per_second_upper_estimate": 0.004047779816696662
     },
     {
       "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": 19288.66643814672,
-      "required_pages_per_second_upper_estimate": 0.004017320542724985
+      "required_token_positions_per_second": 19335.95274421315,
+      "required_pages_per_second_upper_estimate": 0.004027169033254873
     },
     {
       "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": 28932.99965722008,
-      "required_pages_per_second_upper_estimate": 0.0017344891513178641
+      "required_token_positions_per_second": 29003.929116319723,
+      "required_pages_per_second_upper_estimate": 0.0017387412640878746
     },
     {
       "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": 28932.99965722008,
-      "required_pages_per_second_upper_estimate": 0.0022494134605329182
+      "required_token_positions_per_second": 29003.929116319723,
+      "required_pages_per_second_upper_estimate": 0.0022549279139921965
     },
     {
       "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": 28932.99965722008,
-      "required_pages_per_second_upper_estimate": 0.37112145376816713
+      "required_token_positions_per_second": 29003.929116319723,
+      "required_pages_per_second_upper_estimate": 0.37203126071137776
     }
   ],
   "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": 0.08224144300785952,
-    "producer_p95_seconds": 0.18502823900053045,
-    "gpu_update_seconds": 0.9555870572549011,
-    "producer_batches_per_second": 12.159319722836498,
-    "gpu_batches_per_second": 1.046477128805703,
-    "producer_headroom": 11.61928855216681,
+    "producer_batch_seconds": 0.17353317607023655,
+    "producer_p95_seconds": 0.20224193399917567,
+    "gpu_update_seconds": 0.9532501575603155,
+    "producer_batches_per_second": 5.762586858868161,
+    "gpu_batches_per_second": 1.0490425750983696,
+    "producer_headroom": 5.493186831171078,
     "sustainable": true,
-    "buffer_seconds": 60.20198460705877,
+    "buffer_seconds": 60.05475992629988,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 374032.87638595194,
+    "generation_bytes_per_second": 393382.37660202285,
     "upload_bytes_per_second": 4619092.64926371,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.25874823268813,
+    "estimated_full_shards_per_hour": 84.41070054574503,
     "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

@@ -104,3 +104,4 @@
 {"step": 535400, "loss": 0.6516411900520325, "lr": 0.0003, "grad_norm": 0.1685480552530105, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78365211719, "tokens_per_second": 151246.8921404831, "data_wait_s": 2.094999945256859e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 535500, "loss": 0.6496911644935608, "lr": 0.0003, "grad_norm": 0.19610050207521854, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78379842362, "tokens_per_second": 152691.97759033946, "data_wait_s": 1.2108999726478942e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 535600, "loss": 0.6079011559486389, "lr": 0.0003, "grad_norm": 0.15751586357119715, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78394365462, "tokens_per_second": 151797.8055164574, "data_wait_s": 1.3263999790069647e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 535700, "loss": 0.582923948764801, "lr": 0.0003, "grad_norm": 0.1584891579298035, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78408957915, "tokens_per_second": 152132.8347766694, "data_wait_s": 1.3298000340000726e-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 sha256:4adf90247b7e0ca451de793d0a458760af652006eac745d6aab2fbbe89e6a2f7
-size 226239
+oid sha256:e483ef5699895d2ee894a05bd5903c08392e5a3c90a833226eea1b45925c7e0e
+size 228357