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": 75991.51670516684,
-      "required_pages_per_second_upper_estimate": 0.0052572694502749885
+      "required_token_positions_per_second": 77114.65293073296,
+      "required_pages_per_second_upper_estimate": 0.005334970620394738
     },
     {
       "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": 9498.939588145855,
-      "required_pages_per_second_upper_estimate": 0.0039770076451417545
+      "required_token_positions_per_second": 9639.33161634162,
+      "required_pages_per_second_upper_estimate": 0.004035786855628396
     },
     {
       "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": 18997.87917629171,
-      "required_pages_per_second_upper_estimate": 0.003956757224656372
+      "required_token_positions_per_second": 19278.66323268324,
+      "required_pages_per_second_upper_estimate": 0.004015237138829211
     },
     {
       "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": 28496.818764437565,
-      "required_pages_per_second_upper_estimate": 0.0017083407728052124
+      "required_token_positions_per_second": 28917.994849024864,
+      "required_pages_per_second_upper_estimate": 0.0017335896359775764
     },
     {
       "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": 28496.818764437565,
-      "required_pages_per_second_upper_estimate": 0.0022155023146760654
+      "required_token_positions_per_second": 28917.994849024864,
+      "required_pages_per_second_upper_estimate": 0.0022482469026949164
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28496.818764437565,
-      "required_pages_per_second_upper_estimate": 0.36552659361010714
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28917.994849024864,
+      "required_pages_per_second_upper_estimate": 0.3709289881995467
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09569582548422773,
-    "producer_p95_seconds": 0.27176082499863696,
-    "gpu_update_seconds": 0.9702135606274465,
-    "producer_batches_per_second": 10.449776622333612,
-    "gpu_batches_per_second": 1.0307009101720763,
-    "producer_headroom": 10.138514984515744,
+    "producer_batch_seconds": 0.21673038153900848,
+    "producer_p95_seconds": 0.28051862999927835,
+    "gpu_update_seconds": 0.9560828869478933,
+    "producer_batches_per_second": 4.614027774504765,
+    "gpu_batches_per_second": 1.045934420175957,
+    "producer_headroom": 4.411392995106279,
     "sustainable": true,
-    "buffer_seconds": 60.15324075890168,
+    "buffer_seconds": 60.23322187771728,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 349582.5161070806,
-    "upload_bytes_per_second": 4743179.573413656,
+    "generation_bytes_per_second": 392284.7987939317,
+    "upload_bytes_per_second": 4717814.415874457,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.01227009209933,
+    "estimated_full_shards_per_hour": 84.17518589843239,
     "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": 1.2846820170016144,
-      "bytes": 6246650,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.021862665998924,
       "bytes": 6426718,
@@ -562,6 +556,12 @@
       "bytes": 5822650,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 0.9832762490004825,
+      "bytes": 2733834,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -301,3 +301,4 @@
 {"step": 715100, "loss": 0.5293847918510437, "lr": 8e-05, "grad_norm": 0.2190585582182101, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104670736248, "tokens_per_second": 134843.6941909461, "data_wait_s": 6.782000127714127e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 715200, "loss": 0.5196549892425537, "lr": 8e-05, "grad_norm": 0.21338338280575464, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104685388793, "tokens_per_second": 152555.09060464893, "data_wait_s": 1.3106997357681394e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 715300, "loss": 0.5784443020820618, "lr": 8e-05, "grad_norm": 0.2182984198762408, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104700026515, "tokens_per_second": 152363.27701295802, "data_wait_s": 1.5354002243839204e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 715400, "loss": 0.5608413219451904, "lr": 8e-05, "grad_norm": 0.21070986636697486, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104714675976, "tokens_per_second": 152025.3985217592, "data_wait_s": 1.5050998626975343e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:afe86d9a08e4a0ff1acd32fbbe604ef39032efcd3aa865c0411ec7d93fdf7a65
-size 646617
+oid sha256:fe115e3695cba98c31f25dbdb5eb26b44b3edb4a498960337b0c4080fb9aa319
+size 648735