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": 76328.43009047308,
-      "required_pages_per_second_upper_estimate": 0.005280577900017232
+      "required_token_positions_per_second": 77216.28645682354,
+      "required_pages_per_second_upper_estimate": 0.005342001863551955
     },
     {
       "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": 9541.053761309135,
-      "required_pages_per_second_upper_estimate": 0.003994639970001279
+      "required_token_positions_per_second": 9652.035807102942,
+      "required_pages_per_second_upper_estimate": 0.004041105835006491
     },
     {
       "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": 19082.10752261827,
-      "required_pages_per_second_upper_estimate": 0.003974299767945329
+      "required_token_positions_per_second": 19304.071614205885,
+      "required_pages_per_second_upper_estimate": 0.0040205290346362966
     },
     {
       "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": 28623.161283927406,
-      "required_pages_per_second_upper_estimate": 0.0017159148139347712
+      "required_token_positions_per_second": 28956.107421308825,
+      "required_pages_per_second_upper_estimate": 0.0017358744264914716
     },
     {
       "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": 28623.161283927406,
-      "required_pages_per_second_upper_estimate": 0.002225324889844388
+      "required_token_positions_per_second": 28956.107421308825,
+      "required_pages_per_second_upper_estimate": 0.0022512099875505055
     },
     {
       "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": 28623.161283927406,
-      "required_pages_per_second_upper_estimate": 0.36714717979409456
+      "required_token_positions_per_second": 28956.107421308825,
+      "required_pages_per_second_upper_estimate": 0.37141785535471356
     }
   ],
   "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": 0.08021009539062618,
-    "producer_p95_seconds": 0.22054264600046736,
-    "gpu_update_seconds": 0.9659310418491411,
-    "producer_batches_per_second": 12.467258580481202,
-    "gpu_batches_per_second": 1.035270590419828,
-    "producer_headroom": 12.042512069646852,
+    "producer_batch_seconds": 0.4320064028204058,
+    "producer_p95_seconds": 0.23610025300149573,
+    "gpu_update_seconds": 0.9548244727001466,
+    "producer_batches_per_second": 2.3147805066577245,
+    "gpu_batches_per_second": 1.0473129130971073,
+    "producer_headroom": 2.21020907668604,
     "sustainable": true,
-    "buffer_seconds": 60.853655636495894,
+    "buffer_seconds": 60.153941780109236,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 279809.6336833786,
-    "upload_bytes_per_second": 5188023.6488373615,
+    "generation_bytes_per_second": 247472.64858028156,
+    "upload_bytes_per_second": 5119666.668835195,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.04063375354784,
+    "estimated_full_shards_per_hour": 53.101869516909936,
     "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": 1.5193411210002523,
-      "bytes": 12040700,
-      "pages": 35,
-      "shards": 4
-    },
     {
       "seconds": 1.460341175999929,
       "bytes": 11774821,
@@ -562,6 +556,12 @@
       "bytes": 2673736,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 1.4229377800002112,
+      "bytes": 5867572,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -119,3 +119,4 @@
 {"step": 656900, "loss": 0.5579832792282104, "lr": 0.0001, "grad_norm": 0.1867220756938115, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96152450838, "tokens_per_second": 152308.63615351458, "data_wait_s": 8.926999726099893e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 657000, "loss": 0.5376891493797302, "lr": 0.0001, "grad_norm": 0.19578492926159133, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96167131594, "tokens_per_second": 153361.8409411702, "data_wait_s": 1.2873000741819851e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 657100, "loss": 0.5608146786689758, "lr": 0.0001, "grad_norm": 0.1846081286414038, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96181683029, "tokens_per_second": 151656.85201946693, "data_wait_s": 9.3039998319e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 657200, "loss": 0.5723109245300293, "lr": 0.0001, "grad_norm": 0.18259844057913233, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96196414328, "tokens_per_second": 154223.65756405285, "data_wait_s": 1.490099930379074e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bb02455344bb03047313f1f37716f15773a277686ca8b2b02700ad7f242bec3b
-size 258009
+oid sha256:fff30807e9f193522e6fb8eed379098a9bb9f58f6f21e86ef070c24250b7002a
+size 260127