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": 76067.61248420888,
-      "required_pages_per_second_upper_estimate": 0.0065933478585299705
+      "required_token_positions_per_second": 75489.67832872638,
+      "required_pages_per_second_upper_estimate": 0.006543253990693474
     },
     {
       "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": 9508.45156052611,
-      "required_pages_per_second_upper_estimate": 0.003793063491513527
+      "required_token_positions_per_second": 9436.209791090798,
+      "required_pages_per_second_upper_estimate": 0.0037642451695750747
     },
     {
       "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": 19016.90312105222,
-      "required_pages_per_second_upper_estimate": 0.004399225850991449
+      "required_token_positions_per_second": 18872.419582181596,
+      "required_pages_per_second_upper_estimate": 0.004365802127097166
     },
     {
       "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": 28525.354681578327,
-      "required_pages_per_second_upper_estimate": 0.001737044439138515
+      "required_token_positions_per_second": 28308.629373272393,
+      "required_pages_per_second_upper_estimate": 0.001723847004932499
     },
     {
       "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": 28525.354681578327,
-      "required_pages_per_second_upper_estimate": 0.00219596076662353
+      "required_token_positions_per_second": 28308.629373272393,
+      "required_pages_per_second_upper_estimate": 0.0021792766524560887
     },
     {
       "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": 28525.354681578327,
-      "required_pages_per_second_upper_estimate": 0.3541894369243742
+      "required_token_positions_per_second": 28308.629373272393,
+      "required_pages_per_second_upper_estimate": 0.3514984339281621
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.06502661185942316,
-    "producer_p95_seconds": 0.20750523499737028,
-    "gpu_update_seconds": 0.9692429878130517,
-    "producer_batches_per_second": 15.378319297364525,
-    "gpu_batches_per_second": 1.0317330252306975,
-    "producer_headroom": 14.905328143320704,
+    "producer_batch_seconds": 0.27598735062494484,
+    "producer_p95_seconds": 0.2824830879981164,
+    "gpu_update_seconds": 0.9766633218245414,
+    "producer_batches_per_second": 3.6233544680058825,
+    "gpu_batches_per_second": 1.0238942915680118,
+    "producer_headroom": 3.538797410870419,
     "sustainable": true,
-    "buffer_seconds": 60.09306524440921,
+    "buffer_seconds": 60.553125953121565,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 374722.53425824264,
-    "upload_bytes_per_second": 4882971.995982995,
+    "generation_bytes_per_second": 398645.6786203301,
+    "upload_bytes_per_second": 4883286.62227072,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.40673275766811,
+    "estimated_full_shards_per_hour": 85.54008263547351,
     "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": 1.2945012599993788,
-      "bytes": 12063808,
-      "pages": 38,
-      "shards": 4
-    },
     {
       "seconds": 0.8872203939990868,
       "bytes": 6084404,
@@ -562,6 +556,12 @@
       "bytes": 5690031,
       "pages": 26,
       "shards": 2
+    },
+    {
+      "seconds": 1.5474547390003863,
+      "bytes": 13326359,
+      "pages": 36,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -164,3 +164,4 @@
 {"step": 253500, "loss": 0.5892300484701991, "lr": 0.0003, "grad_norm": 0.15466313150434347, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37105693694, "tokens_per_second": 150293.5501557706, "data_wait_s": 1.0577001376077533e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 253600, "loss": 0.610922260209918, "lr": 0.0003, "grad_norm": 0.1601876668024343, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37120402003, "tokens_per_second": 150020.82283779018, "data_wait_s": 1.0567000572336838e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 253700, "loss": 0.645586658269167, "lr": 0.0003, "grad_norm": 0.15923979247640507, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37135003284, "tokens_per_second": 149426.6055320112, "data_wait_s": 1.3652999768964946e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 253800, "loss": 0.5735029503703117, "lr": 0.0003, "grad_norm": 0.14453678746875576, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37149773881, "tokens_per_second": 151450.40581484855, "data_wait_s": 1.007000173558481e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:633697a4acb1e5db077e5b61bf11d56c140479b42ecf5f9dba37679017e3efc7
-size 354102
+oid sha256:451519e7c1d5ed4e37c821480cca712478462fff503ee59baf94e539c68a971c
+size 356220