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": 76660.01445962541,
-      "required_pages_per_second_upper_estimate": 0.005303517676056926
+      "required_token_positions_per_second": 76697.17232830173,
+      "required_pages_per_second_upper_estimate": 0.005306088343630088
     },
     {
       "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": 9582.501807453176,
-      "required_pages_per_second_upper_estimate": 0.0040119934013881624
+      "required_token_positions_per_second": 9587.146541037717,
+      "required_pages_per_second_upper_estimate": 0.004013938054346943
     },
     {
       "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": 19165.003614906353,
-      "required_pages_per_second_upper_estimate": 0.003991564837851972
+      "required_token_positions_per_second": 19174.293082075434,
+      "required_pages_per_second_upper_estimate": 0.003993499588883569
     },
     {
       "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": 28747.505422359533,
-      "required_pages_per_second_upper_estimate": 0.0017233690551712712
+      "required_token_positions_per_second": 28761.43962311315,
+      "required_pages_per_second_upper_estimate": 0.0017242043892301579
     },
     {
       "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": 28747.505422359533,
-      "required_pages_per_second_upper_estimate": 0.0022349920996754223
+      "required_token_positions_per_second": 28761.43962311315,
+      "required_pages_per_second_upper_estimate": 0.0022360754224939445
     },
     {
       "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": 28747.505422359533,
-      "required_pages_per_second_upper_estimate": 0.368742132891568
+      "required_token_positions_per_second": 28761.43962311315,
+      "required_pages_per_second_upper_estimate": 0.3689208658574563
     }
   ],
   "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": 0.11892233741423297,
-    "producer_p95_seconds": 0.31795231599971885,
-    "gpu_update_seconds": 0.9617530145240238,
-    "producer_batches_per_second": 8.40884918463028,
-    "gpu_batches_per_second": 1.0397679912601103,
-    "producer_headroom": 8.087236051996053,
+    "producer_batch_seconds": 0.10777795762538744,
+    "producer_p95_seconds": 0.3142886279965751,
+    "gpu_update_seconds": 0.9612870691556631,
+    "producer_batches_per_second": 9.278335032806808,
+    "gpu_batches_per_second": 1.0402719771091273,
+    "producer_headroom": 8.91914349033117,
     "sustainable": true,
-    "buffer_seconds": 60.590439915013505,
+    "buffer_seconds": 60.56108535680677,
     "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": 254351.59812368272,
-    "upload_bytes_per_second": 4843450.842176247,
+    "generation_bytes_per_second": 305496.1178814102,
+    "upload_bytes_per_second": 4792480.028266157,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 54.57793195517408,
+    "estimated_full_shards_per_hour": 65.55235531169633,
     "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": 1.002878891002183,
-      "bytes": 6148473,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.581660068997735,
       "bytes": 6199462,
@@ -562,6 +556,12 @@
       "bytes": 5997714,
       "pages": 33,
       "shards": 2
+    },
+    {
+      "seconds": 1.251247890999366,
+      "bytes": 2744476,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -390,3 +390,4 @@
 {"step": 759000, "loss": 0.5519813895225525, "lr": 5e-05, "grad_norm": 0.2668603073262277, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111095670229, "tokens_per_second": 149642.5407720595, "data_wait_s": 1.2448006600607187e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 759100, "loss": 0.5515175461769104, "lr": 5e-05, "grad_norm": 0.24843087476895098, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111110173626, "tokens_per_second": 149278.94468124522, "data_wait_s": 1.0984003893099725e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 759200, "loss": 0.5073992609977722, "lr": 5e-05, "grad_norm": 0.23980937587591555, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111124782897, "tokens_per_second": 151043.34325093732, "data_wait_s": 1.4990997442509979e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 759300, "loss": 0.5393655896186829, "lr": 5e-05, "grad_norm": 0.23913106737895645, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111139486457, "tokens_per_second": 152788.36661414208, "data_wait_s": 7.922004442662e-06, "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:bf723f7d74eadd93b47eb8668fa8461517d43246064877a13c9451ff39b0b35a
-size 835902
+oid sha256:2c336173f11cdaafc68f66cd95e50704adcb14cebf284b55749444b5bffd15d6
+size 838020