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": 76608.32019927236,
-      "required_pages_per_second_upper_estimate": 0.005299941347178493
+      "required_token_positions_per_second": 76461.2337084961,
+      "required_pages_per_second_upper_estimate": 0.005289765562458913
     },
     {
       "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": 9576.040024909045,
-      "required_pages_per_second_upper_estimate": 0.004009287987974299
+      "required_token_positions_per_second": 9557.654213562013,
+      "required_pages_per_second_upper_estimate": 0.004001590232702697
     },
     {
       "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": 19152.08004981809,
-      "required_pages_per_second_upper_estimate": 0.003988873200061418
+      "required_token_positions_per_second": 19115.308427124026,
+      "required_pages_per_second_upper_estimate": 0.003981214640787147
     },
     {
       "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": 28728.120074727136,
-      "required_pages_per_second_upper_estimate": 0.001722206933180421
+      "required_token_positions_per_second": 28672.96264068604,
+      "required_pages_per_second_upper_estimate": 0.0017189003292301826
     },
     {
       "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": 28728.120074727136,
-      "required_pages_per_second_upper_estimate": 0.0022334849741641365
+      "required_token_positions_per_second": 28672.96264068604,
+      "required_pages_per_second_upper_estimate": 0.0022291967262793536
     },
     {
       "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": 28728.120074727136,
-      "required_pages_per_second_upper_estimate": 0.3684934784665042
+      "required_token_positions_per_second": 28672.96264068604,
+      "required_pages_per_second_upper_estimate": 0.36778597812606356
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.5142956878435712,
-    "producer_p95_seconds": 0.332353884994518,
-    "gpu_update_seconds": 0.9624019924757504,
-    "producer_batches_per_second": 1.9444067365078925,
-    "gpu_batches_per_second": 1.0390668429805823,
-    "producer_headroom": 1.8713009173984672,
+    "producer_batch_seconds": 0.13802889157022946,
+    "producer_p95_seconds": 0.31979110200336436,
+    "gpu_update_seconds": 0.9642533402100678,
+    "producer_batches_per_second": 7.24486003346044,
+    "gpu_batches_per_second": 1.0370718547701838,
+    "producer_headroom": 6.985880486618653,
     "sustainable": true,
-    "buffer_seconds": 60.63132552597228,
+    "buffer_seconds": 60.74796043323427,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 270219.15426799917,
-    "upload_bytes_per_second": 4879999.152468167,
+    "generation_bytes_per_second": 269989.1271183161,
+    "upload_bytes_per_second": 4898720.865222296,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.98274012594205,
+    "estimated_full_shards_per_hour": 57.933381654378046,
     "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": [
-    {
-      "seconds": 0.9918526799992833,
-      "bytes": 6095255,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.0435619140007475,
       "bytes": 6434451,
@@ -562,6 +556,12 @@
       "bytes": 6238095,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.1314651699940441,
+      "bytes": 8436407,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -380,3 +380,4 @@
 {"step": 758000, "loss": 0.5558410286903381, "lr": 5e-05, "grad_norm": 0.257306425486549, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110949712058, "tokens_per_second": 152647.6719865972, "data_wait_s": 1.3356999261304736e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 758100, "loss": 0.5570091605186462, "lr": 5e-05, "grad_norm": 0.24113663297598134, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110964444119, "tokens_per_second": 152024.26559549387, "data_wait_s": 1.5047000488266349e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 758200, "loss": 0.5575207471847534, "lr": 5e-05, "grad_norm": 0.24727222334973173, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110979052216, "tokens_per_second": 151033.8948268855, "data_wait_s": 9.016999683808535e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 758300, "loss": 0.5599696636199951, "lr": 5e-05, "grad_norm": 0.2452968377925745, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110993613847, "tokens_per_second": 150560.9622968602, "data_wait_s": 1.1952004570048302e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d936f517576b8c77450b4db7db3764b12a9c5e6a74e98af3de56828b018d51cc
-size 814722
+oid sha256:72c93e2362485f8913a05ec4e33b4f3ecaf21383f1d41f38a0005932b1effab6
+size 816840