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": 77064.1556326709,
-      "required_pages_per_second_upper_estimate": 0.00533147710014493
+      "required_token_positions_per_second": 76315.02909814843,
+      "required_pages_per_second_upper_estimate": 0.005279650788273626
     },
     {
       "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": 9633.019454083862,
-      "required_pages_per_second_upper_estimate": 0.004033144085104009
+      "required_token_positions_per_second": 9539.378637268554,
+      "required_pages_per_second_upper_estimate": 0.003993938630546055
     },
     {
       "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": 19266.038908167724,
-      "required_pages_per_second_upper_estimate": 0.0040126078249584545
+      "required_token_positions_per_second": 19078.757274537107,
+      "required_pages_per_second_upper_estimate": 0.003973601999622006
     },
     {
       "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": 28899.058362251584,
-      "required_pages_per_second_upper_estimate": 0.0017324544225098594
+      "required_token_positions_per_second": 28618.135911805664,
+      "required_pages_per_second_upper_estimate": 0.0017156135505493717
     },
     {
       "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": 28899.058362251584,
-      "required_pages_per_second_upper_estimate": 0.0022467746741410967
+      "required_token_positions_per_second": 28618.135911805664,
+      "required_pages_per_second_upper_estimate": 0.0022249341892661987
     },
     {
       "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": 28899.058362251584,
-      "required_pages_per_second_upper_estimate": 0.3706860912796345
+      "required_token_positions_per_second": 28618.135911805664,
+      "required_pages_per_second_upper_estimate": 0.36708271971634104
     }
   ],
   "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": 0.18531478953934766,
-    "producer_p95_seconds": 0.3320143230012036,
-    "gpu_update_seconds": 0.9567093727909924,
-    "producer_batches_per_second": 5.3962233801510555,
-    "gpu_batches_per_second": 1.0452495067365302,
-    "producer_headroom": 5.162617485464405,
+    "producer_batch_seconds": 0.22102456942207027,
+    "producer_p95_seconds": 0.34363604099780787,
+    "gpu_update_seconds": 0.9661006602667835,
+    "producer_batches_per_second": 4.52438388462774,
+    "gpu_batches_per_second": 1.0350888278286192,
+    "producer_headroom": 4.371010258239255,
     "sustainable": true,
-    "buffer_seconds": 60.27269048583253,
+    "buffer_seconds": 60.864341596807364,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 412463.5404227391,
-    "upload_bytes_per_second": 3768809.7161592585,
+    "generation_bytes_per_second": 429244.4598931214,
+    "upload_bytes_per_second": 3837379.2477487396,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 88.50507411491041,
+    "estimated_full_shards_per_hour": 92.10586879344207,
     "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": 1.413499946000229,
-      "bytes": 1206249,
-      "pages": 35,
-      "shards": 1
-    },
     {
       "seconds": 2.568047996999667,
       "bytes": 10001629,
@@ -562,6 +556,12 @@
       "bytes": 5665128,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 2.0745964650050155,
+      "bytes": 11457524,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -338,3 +338,4 @@
 {"step": 833800, "loss": 0.5325481295585632, "lr": 3e-05, "grad_norm": 0.3009460648840372, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122044753261, "tokens_per_second": 149622.43569814428, "data_wait_s": 1.3773002137895674e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 833900, "loss": 0.5314247608184814, "lr": 3e-05, "grad_norm": 0.2714516571475899, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122059201453, "tokens_per_second": 150076.48858682986, "data_wait_s": 1.505599357187748e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 834000, "loss": 0.5625792145729065, "lr": 3e-05, "grad_norm": 0.30661412219508866, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122073826340, "tokens_per_second": 150689.94743833755, "data_wait_s": 1.661400165176019e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 834100, "loss": 0.5463454127311707, "lr": 3e-05, "grad_norm": 0.29014100913848845, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122088582622, "tokens_per_second": 152620.16902439186, "data_wait_s": 1.592100306879729e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7f1a601aa920f624fc84d38b0d21cbc7f81e4ec958fe9d1d435cd9ffb48d9f6b
-size 724983
+oid sha256:4f858ab3c0931c29cb4c58d040ccc0271e8e3fa03af97441a5dbc7867670add7
+size 727101