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": 77216.28645682354,
-      "required_pages_per_second_upper_estimate": 0.005342001863551955
+      "required_token_positions_per_second": 76959.91088843411,
+      "required_pages_per_second_upper_estimate": 0.005324265206857498
     },
     {
       "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": 9652.035807102942,
-      "required_pages_per_second_upper_estimate": 0.004041105835006491
+      "required_token_positions_per_second": 9619.988861054264,
+      "required_pages_per_second_upper_estimate": 0.004027688447912242
     },
     {
       "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": 19304.071614205885,
-      "required_pages_per_second_upper_estimate": 0.0040205290346362966
+      "required_token_positions_per_second": 19239.977722108528,
+      "required_pages_per_second_upper_estimate": 0.004007179967182018
     },
     {
       "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": 28956.107421308825,
-      "required_pages_per_second_upper_estimate": 0.0017358744264914716
+      "required_token_positions_per_second": 28859.96658316279,
+      "required_pages_per_second_upper_estimate": 0.0017301109300431768
     },
     {
       "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": 28956.107421308825,
-      "required_pages_per_second_upper_estimate": 0.0022512099875505055
+      "required_token_positions_per_second": 28859.96658316279,
+      "required_pages_per_second_upper_estimate": 0.002243735460263507
     },
     {
       "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": 28956.107421308825,
-      "required_pages_per_second_upper_estimate": 0.37141785535471356
+      "required_token_positions_per_second": 28859.96658316279,
+      "required_pages_per_second_upper_estimate": 0.3701846639109656
     }
   ],
   "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": 0.4320064028204058,
-    "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,
+    "producer_batch_seconds": 0.10953216847647695,
+    "producer_p95_seconds": 0.2094678550001845,
+    "gpu_update_seconds": 0.9580052672732523,
+    "producer_batches_per_second": 9.129737993042284,
+    "gpu_batches_per_second": 1.0438355969025894,
+    "producer_headroom": 8.746337086159238,
     "sustainable": true,
-    "buffer_seconds": 60.153941780109236,
+    "buffer_seconds": 60.3543318382149,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 247472.64858028156,
-    "upload_bytes_per_second": 5119666.668835195,
+    "generation_bytes_per_second": 248297.2513333754,
+    "upload_bytes_per_second": 5128051.11877952,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 53.101869516909936,
+    "estimated_full_shards_per_hour": 53.27881007195422,
     "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": 1.460341175999929,
-      "bytes": 11774821,
-      "pages": 30,
-      "shards": 4
-    },
     {
       "seconds": 1.2943038749999687,
       "bytes": 2497791,
@@ -562,6 +556,12 @@
       "bytes": 5867572,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.0479602760005946,
+      "bytes": 10355941,
+      "pages": 43,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -120,3 +120,4 @@
 {"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}
+{"step": 657300, "loss": 0.5231503844261169, "lr": 0.0001, "grad_norm": 0.1778715153824788, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96211010820, "tokens_per_second": 153098.66837463455, "data_wait_s": 9.149000106845051e-06, "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:fff30807e9f193522e6fb8eed379098a9bb9f58f6f21e86ef070c24250b7002a
-size 260127
+oid sha256:884d9dcee6014c055530d92063dccbfbeff0eec2e28de5817b5c728f45210dba
+size 262245