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": 75330.5165732158,
-      "required_pages_per_second_upper_estimate": 0.006529458253117598
+      "required_token_positions_per_second": 75731.54024802553,
+      "required_pages_per_second_upper_estimate": 0.0065642179688648935
     },
     {
       "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": 9416.314571651976,
-      "required_pages_per_second_upper_estimate": 0.0037563086690808904
+      "required_token_positions_per_second": 9466.442531003191,
+      "required_pages_per_second_upper_estimate": 0.0037763054615458716
     },
     {
       "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": 18832.62914330395,
-      "required_pages_per_second_upper_estimate": 0.0043565973093506296
+      "required_token_positions_per_second": 18932.885062006382,
+      "required_pages_per_second_upper_estimate": 0.004379789751698531
     },
     {
       "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": 28248.943714955927,
-      "required_pages_per_second_upper_estimate": 0.0017202124614874741
+      "required_token_positions_per_second": 28399.327593009573,
+      "required_pages_per_second_upper_estimate": 0.0017293700506577105
     },
     {
       "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": 28248.943714955927,
-      "required_pages_per_second_upper_estimate": 0.0021746818852584103
+      "required_token_positions_per_second": 28399.327593009573,
+      "required_pages_per_second_upper_estimate": 0.002186258852480198
     },
     {
       "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": 28248.943714955927,
-      "required_pages_per_second_upper_estimate": 0.35075733780692014
+      "required_token_positions_per_second": 28399.327593009573,
+      "required_pages_per_second_upper_estimate": 0.35262460226988307
     }
   ],
   "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": 0.14715815595340587,
-    "producer_p95_seconds": 0.3216775079999934,
-    "gpu_update_seconds": 0.9787268606918648,
-    "producer_batches_per_second": 6.795409969099003,
-    "gpu_batches_per_second": 1.0217355220976536,
-    "producer_headroom": 6.650850266170469,
+    "producer_batch_seconds": 0.5608623446175045,
+    "producer_p95_seconds": 0.366105711000273,
+    "gpu_update_seconds": 0.9735441766869681,
+    "producer_batches_per_second": 1.7829686902620954,
+    "gpu_batches_per_second": 1.0271747537980893,
+    "producer_headroom": 1.7357987856198536,
     "sustainable": true,
-    "buffer_seconds": 60.681065362895616,
+    "buffer_seconds": 60.359738954592025,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 385897.14722497226,
-    "upload_bytes_per_second": 4427871.854131858,
+    "generation_bytes_per_second": 317608.2355236879,
+    "upload_bytes_per_second": 4417748.992388862,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.80454456865192,
+    "estimated_full_shards_per_hour": 68.15133380206088,
     "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": 1.0375558390005608,
-      "bytes": 11763695,
-      "pages": 40,
-      "shards": 3
-    },
     {
       "seconds": 1.2267565980000654,
       "bytes": 917900,
@@ -562,6 +556,12 @@
       "bytes": 6597417,
       "pages": 26,
       "shards": 2
+    },
+    {
+      "seconds": 1.1337901490005606,
+      "bytes": 11234450,
+      "pages": 36,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -278,3 +278,4 @@
 {"step": 264900, "loss": 0.6083609499037266, "lr": 0.0003, "grad_norm": 0.1712524332294461, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38774309209, "tokens_per_second": 148624.64394045892, "data_wait_s": 1.2454001989681274e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 265000, "loss": 0.5934735797345638, "lr": 0.0003, "grad_norm": 0.15487432055418565, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38789095721, "tokens_per_second": 150745.35371688954, "data_wait_s": 1.3623000995721668e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 265100, "loss": 0.5947478972375393, "lr": 0.0003, "grad_norm": 0.16243270863430434, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38803747659, "tokens_per_second": 124333.38661100768, "data_wait_s": 1.1961998097831383e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 265200, "loss": 0.5762580297887325, "lr": 0.0003, "grad_norm": 0.15203765024384844, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38818274096, "tokens_per_second": 148059.59847431353, "data_wait_s": 1.0679999832063913e-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 sha256:7316ca57cf749f6c67cf9baa067e9a39f91b1c73e7d997c92782142159f48c57
-size 597903
+oid sha256:22d9c45d57e8881db66d122ca9cc7de3ee41e92e46c3749ba37d71238fe1bc78
+size 600021