Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76925.35697640284,
-      "required_pages_per_second_upper_estimate": 0.005321874687047052
+      "required_token_positions_per_second": 76439.64704593534,
+      "required_pages_per_second_upper_estimate": 0.005288272147055002
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9615.669622050355,
-      "required_pages_per_second_upper_estimate": 0.004025880072737272
+      "required_token_positions_per_second": 9554.955880741918,
+      "required_pages_per_second_upper_estimate": 0.004000460497098519
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19231.33924410071,
-      "required_pages_per_second_upper_estimate": 0.004005380800024974
+      "required_token_positions_per_second": 19109.911761483836,
+      "required_pages_per_second_upper_estimate": 0.003980090657653939
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28847.008866151064,
-      "required_pages_per_second_upper_estimate": 0.0017293341346936119
+      "required_token_positions_per_second": 28664.867642225752,
+      "required_pages_per_second_upper_estimate": 0.0017184150464328353
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28847.008866151064,
-      "required_pages_per_second_upper_estimate": 0.002242728054760819
+      "required_token_positions_per_second": 28664.867642225752,
+      "required_pages_per_second_upper_estimate": 0.0022285673757552016
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28847.008866151064,
-      "required_pages_per_second_upper_estimate": 0.37001845623005175
+      "required_token_positions_per_second": 28664.867642225752,
+      "required_pages_per_second_upper_estimate": 0.36768214417754713
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.16508650511727296,
-    "producer_p95_seconds": 0.2934068600006867,
-    "gpu_update_seconds": 0.9584355913046508,
-    "producer_batches_per_second": 6.057430310791467,
-    "gpu_batches_per_second": 1.04336692947595,
-    "producer_headroom": 5.8056568017101355,
+    "producer_batch_seconds": 0.09753456125767457,
+    "producer_p95_seconds": 0.2835858359976555,
+    "gpu_update_seconds": 0.964525646693452,
+    "producer_batches_per_second": 10.25277590943502,
+    "gpu_batches_per_second": 1.036779066920781,
+    "producer_headroom": 9.889065314450857,
     "sustainable": true,
-    "buffer_seconds": 60.381442252193004,
+    "buffer_seconds": 60.765115741687474,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 75,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 346261.1168935799,
-    "upload_bytes_per_second": 4811005.652394439,
+    "generation_bytes_per_second": 333008.8541712662,
+    "upload_bytes_per_second": 4854714.749722261,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.29957513909862,
+    "estimated_full_shards_per_hour": 71.45594805577745,
     "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": 1.2564982580006472,
-      "bytes": 6256940,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.1927087240001129,
       "bytes": 6721845,
@@ -562,6 +556,12 @@
       "bytes": 2529037,
       "pages": 34,
       "shards": 2
+    },
+    {
+      "seconds": 1.3205864709998423,
+      "bytes": 10433001,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -312,3 +312,4 @@
 {"step": 791200, "loss": 0.5363965630531311, "lr": 3e-05, "grad_norm": 0.278750405515782, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115808961492, "tokens_per_second": 151780.25353076376, "data_wait_s": 1.0931002179859206e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 791300, "loss": 0.5309306979179382, "lr": 3e-05, "grad_norm": 0.26792184818739084, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115823612893, "tokens_per_second": 151635.26995532692, "data_wait_s": 1.5662000805605203e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 791400, "loss": 0.5172816514968872, "lr": 3e-05, "grad_norm": 0.29014122444539825, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115838278791, "tokens_per_second": 152309.81981993414, "data_wait_s": 1.1657000868581235e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 791500, "loss": 0.5506086349487305, "lr": 3e-05, "grad_norm": 0.28517336271728827, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115853113635, "tokens_per_second": 153999.3772493908, "data_wait_s": 1.79190028575249e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:73a6a568b0334eb765548ea272e463edc4997545ac40051bc9dbe13644108f4f
-size 669915
+oid sha256:5f83734f2161f5f622b751d57654e6c33a0dee1da72b15e0d0032e019c9cdfe7
+size 672033