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": 76600.2272255747,
-      "required_pages_per_second_upper_estimate": 0.0052993814564798025
+      "required_token_positions_per_second": 76757.7280160644,
+      "required_pages_per_second_upper_estimate": 0.005310277726618064
     },
     {
       "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": 9575.028403196837,
-      "required_pages_per_second_upper_estimate": 0.004008864443088461
+      "required_token_positions_per_second": 9594.71600200805,
+      "required_pages_per_second_upper_estimate": 0.004017107229586902
     },
     {
       "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": 19150.056806393673,
-      "required_pages_per_second_upper_estimate": 0.003988451811812629
+      "required_token_positions_per_second": 19189.4320040161,
+      "required_pages_per_second_upper_estimate": 0.003996652627083591
     },
     {
       "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": 28725.08520959051,
-      "required_pages_per_second_upper_estimate": 0.0017220249976494508
+      "required_token_positions_per_second": 28784.14800602415,
+      "required_pages_per_second_upper_estimate": 0.0017255657221119792
     },
     {
       "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": 28725.08520959051,
-      "required_pages_per_second_upper_estimate": 0.002233249026748725
+      "required_token_positions_per_second": 28784.14800602415,
+      "required_pages_per_second_upper_estimate": 0.0022378408993816543
     },
     {
       "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": 28725.08520959051,
-      "required_pages_per_second_upper_estimate": 0.3684545504751159
+      "required_token_positions_per_second": 28784.14800602415,
+      "required_pages_per_second_upper_estimate": 0.36921214461107665
     }
   ],
   "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": 0.4761786220549311,
-    "producer_p95_seconds": 0.30748071600100957,
-    "gpu_update_seconds": 0.96250367225261,
-    "producer_batches_per_second": 2.1000522780391466,
-    "gpu_batches_per_second": 1.0389570749996568,
-    "producer_headroom": 2.021308029535138,
+    "producer_batch_seconds": 0.08770319043730979,
+    "producer_p95_seconds": 0.3089865079964511,
+    "gpu_update_seconds": 0.9605286907993119,
+    "producer_batches_per_second": 11.402093755241433,
+    "gpu_batches_per_second": 1.0410933161901095,
+    "producer_headroom": 10.952038187093063,
     "sustainable": true,
-    "buffer_seconds": 60.63773135191443,
+    "buffer_seconds": 60.51330752035665,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 319611.5540382315,
-    "upload_bytes_per_second": 5071335.461176242,
+    "generation_bytes_per_second": 319557.85916930385,
+    "upload_bytes_per_second": 5153095.697937076,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.58119932041367,
+    "estimated_full_shards_per_hour": 68.56967765149437,
     "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": 1.0317238650022773,
-      "bytes": 2334152,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.198212211002101,
       "bytes": 5956669,
@@ -562,6 +556,12 @@
       "bytes": 6532250,
       "pages": 41,
       "shards": 3
+    },
+    {
+      "seconds": 1.5230046290016617,
+      "bytes": 11668778,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -365,3 +365,4 @@
 {"step": 756500, "loss": 0.5631418824195862, "lr": 5e-05, "grad_norm": 0.2474035999962852, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110730424511, "tokens_per_second": 150820.3350710897, "data_wait_s": 1.3278993719723076e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 756600, "loss": 0.534017026424408, "lr": 5e-05, "grad_norm": 0.23299655297612062, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110744899729, "tokens_per_second": 142018.43787473597, "data_wait_s": 1.5256002370733768e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 756700, "loss": 0.499985933303833, "lr": 5e-05, "grad_norm": 0.23385300021185157, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110759322182, "tokens_per_second": 149120.42494103522, "data_wait_s": 1.2472002708818763e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 756800, "loss": 0.5425806045532227, "lr": 5e-05, "grad_norm": 0.24550936957946895, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110773884834, "tokens_per_second": 150261.21545031416, "data_wait_s": 1.2894997780676931e-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 sha256:72b9fb5d5a7443041913ecbf0cef7f0daeacaf49e67ce4e806b7f44519239818
-size 782952
+oid sha256:e71c660b89088eb6add8b6ed9058ad178fc9460f9ca9c3c43010356a742b449e
+size 785070