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": 75952.44852296314,
-      "required_pages_per_second_upper_estimate": 0.005254566622779465
+      "required_token_positions_per_second": 75967.31794180794,
+      "required_pages_per_second_upper_estimate": 0.005255595323676953
     },
     {
       "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": 9494.056065370392,
-      "required_pages_per_second_upper_estimate": 0.0039749630161352205
+      "required_token_positions_per_second": 9495.914742725992,
+      "required_pages_per_second_upper_estimate": 0.003975741205530413
     },
     {
       "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": 18988.112130740785,
-      "required_pages_per_second_upper_estimate": 0.003954723006642426
+      "required_token_positions_per_second": 18991.829485451985,
+      "required_pages_per_second_upper_estimate": 0.003955497233595532
     },
     {
       "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": 28482.16819611118,
-      "required_pages_per_second_upper_estimate": 0.001707462493604168
+      "required_token_positions_per_second": 28487.744228177973,
+      "required_pages_per_second_upper_estimate": 0.0017077967682124645
     },
     {
       "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": 28482.16819611118,
-      "required_pages_per_second_upper_estimate": 0.002214363297429729
+      "required_token_positions_per_second": 28487.744228177973,
+      "required_pages_per_second_upper_estimate": 0.002214796809396549
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28482.16819611118,
-      "required_pages_per_second_upper_estimate": 0.365338671850171
+      "required_token_positions_per_second": 28487.744228177973,
+      "required_pages_per_second_upper_estimate": 0.36541019520244705
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.20598060080442337,
-    "producer_p95_seconds": 0.3176107379986206,
-    "gpu_update_seconds": 0.9707126160351419,
-    "producer_batches_per_second": 4.854826115151933,
-    "gpu_batches_per_second": 1.030171014037586,
-    "producer_headroom": 4.712640958634859,
+    "producer_batch_seconds": 0.24069697325774086,
+    "producer_p95_seconds": 0.3421043490016018,
+    "gpu_update_seconds": 0.9705226141651693,
+    "producer_batches_per_second": 4.1546014744821465,
+    "gpu_batches_per_second": 1.03037269343815,
+    "producer_headroom": 4.03213468382888,
     "sustainable": true,
-    "buffer_seconds": 60.184182194178796,
+    "buffer_seconds": 60.172402078240495,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 343121.4969128226,
+    "generation_bytes_per_second": 369371.7090103406,
     "upload_bytes_per_second": 4604400.782809922,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.6258857778407,
+    "estimated_full_shards_per_hour": 79.2585702203051,
     "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-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -416,3 +416,4 @@
 {"step": 761600, "loss": 0.5514082312583923, "lr": 5e-05, "grad_norm": 0.253750221363891, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111475755654, "tokens_per_second": 149425.7521523962, "data_wait_s": 1.4593999367207289e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 761700, "loss": 0.504923403263092, "lr": 5e-05, "grad_norm": 0.23508749085410036, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111490271322, "tokens_per_second": 149424.3890953439, "data_wait_s": 9.561001206748188e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 761800, "loss": 0.5890706777572632, "lr": 5e-05, "grad_norm": 0.27169946887613655, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111505032369, "tokens_per_second": 152226.03419675006, "data_wait_s": 1.207699824590236e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 761900, "loss": 0.5406587719917297, "lr": 5e-05, "grad_norm": 0.2462840110372524, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111519760426, "tokens_per_second": 152427.16208617773, "data_wait_s": 1.0782998288050294e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4d8b1963fde5ee711d7e0e026d71be1b36f1871bda612e1d77e5e0e46c1776bf
-size 891753
+oid sha256:5137c9658fee19211fc301269988f7321408eb9404a9bc8669bd6d0969c74e91
+size 893871