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": 75627.32412712398,
-      "required_pages_per_second_upper_estimate": 0.006555184779638491
+      "required_token_positions_per_second": 75897.34787940634,
+      "required_pages_per_second_upper_estimate": 0.006578589754117382
     },
     {
       "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": 9453.415515890498,
-      "required_pages_per_second_upper_estimate": 0.0037711087904461855
+      "required_token_positions_per_second": 9487.168484925793,
+      "required_pages_per_second_upper_estimate": 0.0037845733544462235
     },
     {
       "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": 18906.831031780996,
-      "required_pages_per_second_upper_estimate": 0.0043737626103411585
+      "required_token_positions_per_second": 18974.336969851585,
+      "required_pages_per_second_upper_estimate": 0.004389378921049857
     },
     {
       "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": 28360.246547671497,
-      "required_pages_per_second_upper_estimate": 0.0017269902200390199
+      "required_token_positions_per_second": 28461.505454777376,
+      "required_pages_per_second_upper_estimate": 0.0017331563562173406
     },
     {
       "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": 28360.246547671497,
-      "required_pages_per_second_upper_estimate": 0.0021832502854268153
+      "required_token_positions_per_second": 28461.505454777376,
+      "required_pages_per_second_upper_estimate": 0.002191045476398421
     },
     {
       "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": 28360.246547671497,
-      "required_pages_per_second_upper_estimate": 0.35213934648262907
+      "required_token_positions_per_second": 28461.505454777376,
+      "required_pages_per_second_upper_estimate": 0.3533966432171223
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.16837482361677303,
-    "producer_p95_seconds": 0.5921860819944413,
-    "gpu_update_seconds": 0.9748857420377407,
-    "producer_batches_per_second": 5.939130200819304,
-    "gpu_batches_per_second": 1.0257612321929794,
-    "producer_headroom": 5.789973352884482,
+    "producer_batch_seconds": 0.12989549892171226,
+    "producer_p95_seconds": 0.5872306089950143,
+    "gpu_update_seconds": 0.9714173427660051,
+    "producer_batches_per_second": 7.698496162693812,
+    "gpu_batches_per_second": 1.0294236637289271,
+    "producer_headroom": 7.47845268565831,
     "sustainable": true,
-    "buffer_seconds": 60.44291600633992,
+    "buffer_seconds": 60.227875251492314,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 363332.71122008836,
+    "generation_bytes_per_second": 363180.0047451357,
     "upload_bytes_per_second": 4833264.740068988,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.96274187519062,
+    "estimated_full_shards_per_hour": 77.92997462049058,
     "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-v1/1790563279934888633-82973089/train.jsonl

@@ -399,3 +399,4 @@
 {"step": 194900, "loss": 0.6080953255295753, "lr": 0.0003, "grad_norm": 0.16244985294831388, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28525593721, "tokens_per_second": 150274.05671662698, "data_wait_s": 1.1323994840495288e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 195000, "loss": 0.587149441242218, "lr": 0.0003, "grad_norm": 0.14609337480068943, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28540343020, "tokens_per_second": 150961.18382806727, "data_wait_s": 1.727200287859887e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 195100, "loss": 0.5759536698460579, "lr": 0.0003, "grad_norm": 0.1433815500195194, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28555098636, "tokens_per_second": 135442.66935053776, "data_wait_s": 9.759001841302961e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 195200, "loss": 0.611761312931776, "lr": 0.0003, "grad_norm": 0.13731755191137826, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28569686874, "tokens_per_second": 148886.4980394214, "data_wait_s": 1.1459997040219605e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:de5c69881805e6b85a6877fcb4735740247b2b15ded614b964859a1399c8e8e8
-size 855747
+oid sha256:957b562bc43e3dad74bc39305131f9d30c94b3c37c9db7b9ec3987a7531190bf
+size 857865