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": 76290.06908604302,
-      "required_pages_per_second_upper_estimate": 0.0052779239967209665
+      "required_token_positions_per_second": 77318.89232175509,
+      "required_pages_per_second_upper_estimate": 0.005349100375366326
     },
     {
       "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": 9536.258635755377,
-      "required_pages_per_second_upper_estimate": 0.0039926323510655285
+      "required_token_positions_per_second": 9664.861540219386,
+      "required_pages_per_second_upper_estimate": 0.004046475701630582
     },
     {
       "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": 19072.517271510755,
-      "required_pages_per_second_upper_estimate": 0.0039723023715515626
+      "required_token_positions_per_second": 19329.723080438773,
+      "required_pages_per_second_upper_estimate": 0.0040258715585779516
     },
     {
       "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": 28608.775907266132,
-      "required_pages_per_second_upper_estimate": 0.001715052432569126
+      "required_token_positions_per_second": 28994.58462065816,
+      "required_pages_per_second_upper_estimate": 0.0017381810758411826
     },
     {
       "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": 28608.775907266132,
-      "required_pages_per_second_upper_estimate": 0.0022242064900835586
+      "required_token_positions_per_second": 28994.58462065816,
+      "required_pages_per_second_upper_estimate": 0.0022542014205566025
     },
     {
       "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": 28608.775907266132,
-      "required_pages_per_second_upper_estimate": 0.36696265962809776
+      "required_token_positions_per_second": 28994.58462065816,
+      "required_pages_per_second_upper_estimate": 0.37191139955436897
     }
   ],
   "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": 0.14257040306259228,
-    "producer_p95_seconds": 0.3073953570019512,
-    "gpu_update_seconds": 0.9664167418284363,
-    "producer_batches_per_second": 7.014078508012444,
-    "gpu_batches_per_second": 1.0347502859977622,
-    "producer_headroom": 6.778522898642246,
+    "producer_batch_seconds": 0.08133040228139521,
+    "producer_p95_seconds": 0.302603428997827,
+    "gpu_update_seconds": 0.9535573749968903,
+    "producer_batches_per_second": 12.295525067490729,
+    "gpu_batches_per_second": 1.0487045942078328,
+    "producer_headroom": 11.724488607564922,
     "sustainable": true,
-    "buffer_seconds": 60.88425473519148,
+    "buffer_seconds": 60.07411462480409,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 358066.11235102086,
+    "generation_bytes_per_second": 348993.2572749352,
     "upload_bytes_per_second": 5109420.410021223,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.83265235803574,
+    "estimated_full_shards_per_hour": 74.88582886396448,
     "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-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -284,3 +284,4 @@
 {"step": 953400, "loss": 0.5547279119491577, "lr": 3e-05, "grad_norm": 0.3430259935499054, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139551962961, "tokens_per_second": 151408.32961208528, "data_wait_s": 8.521998097421601e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 953500, "loss": 0.568813681602478, "lr": 3e-05, "grad_norm": 0.35661222057761044, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139566538487, "tokens_per_second": 151904.02677224466, "data_wait_s": 1.3524000678444281e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 953600, "loss": 0.5567885041236877, "lr": 3e-05, "grad_norm": 0.33955340563023323, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139581171073, "tokens_per_second": 151886.19815135363, "data_wait_s": 1.3534001482184976e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 953700, "loss": 0.5545731782913208, "lr": 3e-05, "grad_norm": 0.3466795428070545, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139595789915, "tokens_per_second": 151432.76228041694, "data_wait_s": 1.2021999282296747e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:290218e9f0942b5ce53193dcc8bbe5032009a59e61ed75625b3f8312e4cbb946
-size 609828
+oid sha256:315265db1fa6ef941e504ff7cdd6a631d64497577a59c6d472e5d6802172c975
+size 611946