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": 77250.05065003202,
-      "required_pages_per_second_upper_estimate": 0.005344337748781839
+      "required_token_positions_per_second": 77244.09759889393,
+      "required_pages_per_second_upper_estimate": 0.005343925902891124
     },
     {
       "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": 9656.256331254002,
-      "required_pages_per_second_upper_estimate": 0.004042872880333973
+      "required_token_positions_per_second": 9655.512199861741,
+      "required_pages_per_second_upper_estimate": 0.004042561328059264
     },
     {
       "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": 19312.512662508005,
-      "required_pages_per_second_upper_estimate": 0.004022287082392215
+      "required_token_positions_per_second": 19311.024399723483,
+      "required_pages_per_second_upper_estimate": 0.004021977116502328
     },
     {
       "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": 28968.768993762005,
-      "required_pages_per_second_upper_estimate": 0.001736633468426938
+      "required_token_positions_per_second": 28966.536599585223,
+      "required_pages_per_second_upper_estimate": 0.0017364996398047084
     },
     {
       "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": 28968.768993762005,
-      "required_pages_per_second_upper_estimate": 0.002252194369116369
+      "required_token_positions_per_second": 28966.536599585223,
+      "required_pages_per_second_upper_estimate": 0.002252020810288388
     },
     {
       "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": 28968.768993762005,
-      "required_pages_per_second_upper_estimate": 0.3715802644112057
+      "required_token_positions_per_second": 28966.536599585223,
+      "required_pages_per_second_upper_estimate": 0.371551629655664
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,17 +158,17 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.05521727650780939,
-    "producer_p95_seconds": 0.11915372700013904,
-    "gpu_update_seconds": 0.9544071412200354,
-    "producer_batches_per_second": 18.110273871594696,
-    "gpu_batches_per_second": 1.047770869276693,
-    "producer_headroom": 17.284574712500596,
+    "producer_batch_seconds": 0.04334722902342847,
+    "producer_p95_seconds": 0.12166237700012061,
+    "gpu_update_seconds": 0.9544806955069629,
+    "producer_batches_per_second": 23.069525377493367,
+    "gpu_batches_per_second": 1.0476901258530535,
+    "producer_headroom": 22.0194166273254,
     "sustainable": true,
-    "buffer_seconds": 60.12764989686223,
+    "buffer_seconds": 60.13228381693866,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 58,
+    "upload_pages_per_commit": 63,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

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

@@ -7,3 +7,4 @@
 {"step": 760700, "loss": 0.44909876585006714, "lr": 3e-05, "grad_norm": 0.24061763182040033, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111344364855, "tokens_per_second": 154546.64200430753, "data_wait_s": 1.3811999906465644e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 760800, "loss": 0.5324825048446655, "lr": 3e-05, "grad_norm": 0.23670901506295858, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111358936214, "tokens_per_second": 152576.84177390873, "data_wait_s": 1.2162999837528332e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 760900, "loss": 0.5496805310249329, "lr": 3e-05, "grad_norm": 0.36506922309322815, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111373484123, "tokens_per_second": 152380.35194150993, "data_wait_s": 1.4608999890697305e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 761000, "loss": 0.5713914632797241, "lr": 3e-05, "grad_norm": 0.24474886554669245, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111387964808, "tokens_per_second": 151660.63421319184, "data_wait_s": 1.22909998481191e-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:fd42f28641c0dc82f64cc63f2b962ccc3c2e1e43b10cc61fc50705a75d5052cd
-size 19227
+oid sha256:e3349a5a89813c64650efd5282ba9cc6f9f8243896f24f5a35c01ae8c640b26e
+size 21345