Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75755.50772488576,
-      "required_pages_per_second_upper_estimate": 0.0065662954090141236
+      "required_token_positions_per_second": 75801.79695830421,
+      "required_pages_per_second_upper_estimate": 0.006570307642447846
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9469.43846561072,
-      "required_pages_per_second_upper_estimate": 0.003777500584654029
+      "required_token_positions_per_second": 9475.224619788027,
+      "required_pages_per_second_upper_estimate": 0.003779808768066071
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18938.87693122144,
-      "required_pages_per_second_upper_estimate": 0.0043811758651881375
+      "required_token_positions_per_second": 18950.449239576054,
+      "required_pages_per_second_upper_estimate": 0.004383852915060307
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28408.315396832157,
-      "required_pages_per_second_upper_estimate": 0.0017299173607551435
+      "required_token_positions_per_second": 28425.673859364084,
+      "required_pages_per_second_upper_estimate": 0.0017309744000505234
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28408.315396832157,
-      "required_pages_per_second_upper_estimate": 0.0021869507584982937
+      "required_token_positions_per_second": 28425.673859364084,
+      "required_pages_per_second_upper_estimate": 0.0021882870609953014
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28408.315396832157,
-      "required_pages_per_second_upper_estimate": 0.3527362007131152
+      "required_token_positions_per_second": 28425.673859364084,
+      "required_pages_per_second_upper_estimate": 0.3529517347227248
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.2198642464765257,
-    "producer_p95_seconds": 0.18524565100051404,
-    "gpu_update_seconds": 0.9732361674315634,
-    "producer_batches_per_second": 4.548261102137711,
-    "gpu_batches_per_second": 1.0274998335081076,
-    "producer_headroom": 4.426532203522564,
+    "producer_batch_seconds": 0.05076408121094289,
+    "producer_p95_seconds": 0.18196632899980614,
+    "gpu_update_seconds": 0.9726418496457948,
+    "producer_batches_per_second": 19.698967776933515,
+    "gpu_batches_per_second": 1.0281276714179717,
+    "producer_headroom": 19.160040454669524,
     "sustainable": true,
-    "buffer_seconds": 60.34064238075693,
+    "buffer_seconds": 60.30379467803928,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 360240.6442909522,
-    "upload_bytes_per_second": 5245685.810701095,
+    "generation_bytes_per_second": 360146.53058725555,
+    "upload_bytes_per_second": 5427891.649648744,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.29925629183221,
+    "estimated_full_shards_per_hour": 77.2790616818738,
     "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": [
@@ -454,6 +454,12 @@
       "bytes": 647903,
       "pages": 20,
       "shards": 1
+    },
+    {
+      "seconds": 1.4357455670005947,
+      "bytes": 17610080,
+      "pages": 43,
+      "shards": 5
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -55,3 +55,4 @@
 {"step": 320500, "loss": 0.6146031823009253, "lr": 0.0003, "grad_norm": 0.16981394066016361, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46912323849, "tokens_per_second": 131657.32953579738, "data_wait_s": 1.104199964174768e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 320600, "loss": 0.5646807532757521, "lr": 0.0003, "grad_norm": 0.14598938220835858, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46926873367, "tokens_per_second": 148443.65597854287, "data_wait_s": 1.2337000043771695e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 320700, "loss": 0.5956653617322445, "lr": 0.0003, "grad_norm": 0.1756550220108485, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46941285266, "tokens_per_second": 146610.39086905512, "data_wait_s": 1.580099979037186e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 320800, "loss": 0.5841874033212662, "lr": 0.0003, "grad_norm": 0.1514964677834582, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46955936745, "tokens_per_second": 148588.1178477004, "data_wait_s": 1.2708999747701455e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f0e4a02e15e79704d8c2e3c510e519d4f0f8326c390b5ca93712dc2bc3d07a84
-size 121674
+oid sha256:45174cdc70d3ebcf811079ff854765a009006b4ffa5bbf93fa7e42ce266b2fcb
+size 123792