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": 75960.78953772073,
-      "required_pages_per_second_upper_estimate": 0.006584088716268661
+      "required_token_positions_per_second": 75951.3833665632,
+      "required_pages_per_second_upper_estimate": 0.006583273413192449
     },
     {
       "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": 9495.098692215091,
-      "required_pages_per_second_upper_estimate": 0.00378773683270109
+      "required_token_positions_per_second": 9493.9229208204,
+      "required_pages_per_second_upper_estimate": 0.003787267799912398
     },
     {
       "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": 18990.197384430183,
-      "required_pages_per_second_upper_estimate": 0.004393047948828853
+      "required_token_positions_per_second": 18987.8458416408,
+      "required_pages_per_second_upper_estimate": 0.004392503960790268
     },
     {
       "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": 28485.296076645274,
-      "required_pages_per_second_upper_estimate": 0.0017346050802693493
+      "required_token_positions_per_second": 28481.7687624612,
+      "required_pages_per_second_upper_estimate": 0.0017343902853419265
     },
     {
       "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": 28485.296076645274,
-      "required_pages_per_second_upper_estimate": 0.002192876944326468
+      "required_token_positions_per_second": 28481.7687624612,
+      "required_pages_per_second_upper_estimate": 0.0021926054019163477
     },
     {
       "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": 28485.296076645274,
-      "required_pages_per_second_upper_estimate": 0.3536920431186321
+      "required_token_positions_per_second": 28481.7687624612,
+      "required_pages_per_second_upper_estimate": 0.3536482456816271
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.11523102538285457,
+    "producer_batch_seconds": 0.11785942181256814,
     "producer_p95_seconds": 0.3124006370017014,
-    "gpu_update_seconds": 0.970606025143907,
-    "producer_batches_per_second": 8.678218358966298,
-    "gpu_batches_per_second": 1.0302841462907,
-    "producer_headroom": 8.423131026727159,
+    "gpu_update_seconds": 0.9707262294903503,
+    "producer_batches_per_second": 8.484684419971957,
+    "gpu_batches_per_second": 1.030156566929297,
+    "producer_headroom": 8.236305715414897,
     "sustainable": true,
-    "buffer_seconds": 60.177573558922234,
+    "buffer_seconds": 60.18502622840172,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 283906.5385168827,
+    "generation_bytes_per_second": 283881.7911544228,
     "upload_bytes_per_second": 5213977.019993049,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.91973415975438,
+    "estimated_full_shards_per_hour": 60.9144239518596,
     "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": [

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -233,3 +233,4 @@
 {"step": 378300, "loss": 0.557722071185708, "lr": 0.0003, "grad_norm": 0.163479854615462, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55372456151, "tokens_per_second": 150226.34467800736, "data_wait_s": 1.0760002624010667e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 378400, "loss": 0.5520606003701687, "lr": 0.0003, "grad_norm": 0.1735362186521396, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55386921035, "tokens_per_second": 122804.24020786358, "data_wait_s": 1.145099668065086e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 378500, "loss": 0.6116420533508062, "lr": 0.0003, "grad_norm": 0.17059345758723035, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55401541945, "tokens_per_second": 148714.45944499297, "data_wait_s": 8.55699909152463e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 378600, "loss": 0.5808735452592373, "lr": 0.0003, "grad_norm": 0.14859657668266757, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55416197094, "tokens_per_second": 132678.88405871144, "data_wait_s": 1.4684999769087881e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b465a4c1343bf9422c66aaee8f2d1558d6c748e5c070f95e424ba1916bdd01b1
-size 501027
+oid sha256:4caf1853726df9f7be915e40630381cfa9dc4a94497ea558118391585e2f8847
+size 503145