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": 72215.74480393795,
-      "required_pages_per_second_upper_estimate": 0.00625947773047347
+      "required_token_positions_per_second": 76346.25695964231,
+      "required_pages_per_second_upper_estimate": 0.006617500055442574
     },
     {
       "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": 9026.968100492244,
-      "required_pages_per_second_upper_estimate": 0.0036009925404867737
+      "required_token_positions_per_second": 9543.282119955289,
+      "required_pages_per_second_upper_estimate": 0.0038069579224330975
     },
     {
       "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": 18053.936200984488,
-      "required_pages_per_second_upper_estimate": 0.004176460401672738
+      "required_token_positions_per_second": 19086.564239910578,
+      "required_pages_per_second_upper_estimate": 0.004415340724845511
     },
     {
       "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": 27080.904301476734,
-      "required_pages_per_second_upper_estimate": 0.0016490849894357793
+      "required_token_positions_per_second": 28629.846359865867,
+      "required_pages_per_second_upper_estimate": 0.0017434074341207593
     },
     {
       "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": 27080.904301476734,
-      "required_pages_per_second_upper_estimate": 0.0020847629778687426
+      "required_token_positions_per_second": 28629.846359865867,
+      "required_pages_per_second_upper_estimate": 0.0022040048252695834
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 27080.904301476734,
-      "required_pages_per_second_upper_estimate": 0.33625419746795554
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28629.846359865867,
+      "required_pages_per_second_upper_estimate": 0.3554868738575545
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,20 +155,20 @@
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 59,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.04798892122857762,
-    "producer_p95_seconds": 0.13507072599998082,
-    "gpu_update_seconds": 1.0209407962234238,
-    "producer_batches_per_second": 20.838142937968264,
-    "gpu_batches_per_second": 0.9794887261818842,
-    "producer_headroom": 21.274510242906835,
+    "producer_batch_seconds": 0.05906009539063106,
+    "producer_p95_seconds": 0.13741721899998538,
+    "gpu_update_seconds": 0.9657054967209937,
+    "producer_batches_per_second": 16.931906279288434,
+    "gpu_batches_per_second": 1.0355123828076485,
+    "producer_headroom": 16.35123496387355,
     "sustainable": true,
-    "buffer_seconds": 60.235506977182006,
+    "buffer_seconds": 60.839446293422604,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 25,
+    "upload_pages_per_commit": 36,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1790696019647648547-3fdea0a8/train.jsonl

@@ -1,2 +1,3 @@
 {"step": 275100, "loss": 0.5665341280400753, "lr": 0.0003, "grad_norm": 0.15310209458591492, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40266125057, "tokens_per_second": 49579.57407153421, "data_wait_s": 1.2700999945991498e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 275200, "loss": 0.5517940912395716, "lr": 0.0003, "grad_norm": 0.1433728997387089, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40280693026, "tokens_per_second": 149900.41362321944, "data_wait_s": 1.509800017629459e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 275300, "loss": 0.5715921688824892, "lr": 0.0003, "grad_norm": 0.13326310819662313, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40295403551, "tokens_per_second": 151598.20565482703, "data_wait_s": 1.600599989615148e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790696019.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.22407.0.1790696019647648547-3fdea0a8

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9be5d261245c51c156c511fcf5d0960372b009206dbb64f9836e7aaeae28839f
-size 4401
+oid sha256:23adbba35e891f6eb4db3b04b22a021d9ce327828eb0d34fff17190a0b0142b3
+size 6519