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": 76558.29019277034,
-      "required_pages_per_second_upper_estimate": 0.005296480155243075
+      "required_token_positions_per_second": 76901.79793107293,
+      "required_pages_per_second_upper_estimate": 0.005320244817626612
     },
     {
       "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": 9569.786274096292,
-      "required_pages_per_second_upper_estimate": 0.004006669673102166
+      "required_token_positions_per_second": 9612.724741384116,
+      "required_pages_per_second_upper_estimate": 0.004024647112698418
     },
     {
       "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": 19139.572548192584,
-      "required_pages_per_second_upper_estimate": 0.003986268217317824
+      "required_token_positions_per_second": 19225.449482768232,
+      "required_pages_per_second_upper_estimate": 0.0040041541180628705
     },
     {
       "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": 28709.358822288876,
-      "required_pages_per_second_upper_estimate": 0.001721082224743521
+      "required_token_positions_per_second": 28838.17422415235,
+      "required_pages_per_second_upper_estimate": 0.0017288045113955063
     },
     {
       "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": 28709.358822288876,
-      "required_pages_per_second_upper_estimate": 0.00223202636930909
+      "required_token_positions_per_second": 28838.17422415235,
+      "required_pages_per_second_upper_estimate": 0.0022420412002049023
     },
     {
       "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": 28709.358822288876,
-      "required_pages_per_second_upper_estimate": 0.3682528292644896
+      "required_token_positions_per_second": 28838.17422415235,
+      "required_pages_per_second_upper_estimate": 0.3699051349283918
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.28822229649215103,
-    "producer_p95_seconds": 0.1862483169998086,
-    "gpu_update_seconds": 0.9630309116668647,
-    "producer_batches_per_second": 3.469544210044251,
-    "gpu_batches_per_second": 1.0383882675885734,
-    "producer_headroom": 3.341278323667407,
+    "producer_batch_seconds": 0.05265844196873104,
+    "producer_p95_seconds": 0.17604723399927025,
+    "gpu_update_seconds": 0.9587292102855955,
+    "producer_batches_per_second": 18.990307396367843,
+    "gpu_batches_per_second": 1.043047389473103,
+    "producer_headroom": 18.206562413200448,
     "sustainable": true,
-    "buffer_seconds": 60.67094743501248,
+    "buffer_seconds": 60.39994024799252,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 374033.03881379653,
-    "upload_bytes_per_second": 4888539.254230374,
+    "generation_bytes_per_second": 374382.28177509294,
+    "upload_bytes_per_second": 4961836.740156561,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.25878308592245,
+    "estimated_full_shards_per_hour": 80.33372249545661,
     "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": [
@@ -556,6 +556,12 @@
       "bytes": 7087614,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.3239770070003942,
+      "bytes": 12583619,
+      "pages": 38,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -75,3 +75,4 @@
 {"step": 404600, "loss": 0.5431584119796753, "lr": 0.0003, "grad_norm": 0.15124628593235345, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59222939708, "tokens_per_second": 152906.70188366118, "data_wait_s": 9.297999895352405e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 404700, "loss": 0.635788083076477, "lr": 0.0003, "grad_norm": 0.1780795452564756, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59237684063, "tokens_per_second": 153293.41920650937, "data_wait_s": 7.6200003604753874e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 404800, "loss": 0.624845027923584, "lr": 0.0003, "grad_norm": 0.16645235871608838, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59252363994, "tokens_per_second": 152852.3598016883, "data_wait_s": 1.3953000234323554e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 404900, "loss": 0.5621156692504883, "lr": 0.0003, "grad_norm": 0.1732311903236872, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59267149727, "tokens_per_second": 154028.61356223238, "data_wait_s": 7.122999704733957e-06, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:15a7022c60f7b822472571fce6762b054dad793ef7304929c3af32d6233baab9
-size 164034
+oid sha256:6ddbeecfe1c4b9b0ea136ec51a8e53e58eb4ca57117af26634c6d96aaf433475
+size 166152