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": 77014.46467561861,
-      "required_pages_per_second_upper_estimate": 0.005328039364437148
+      "required_token_positions_per_second": 76466.85908460924,
+      "required_pages_per_second_upper_estimate": 0.005290154739031091
     },
     {
       "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": 9626.808084452327,
-      "required_pages_per_second_upper_estimate": 0.004030543514347433
+      "required_token_positions_per_second": 9558.357385576155,
+      "required_pages_per_second_upper_estimate": 0.004001884636141116
     },
     {
       "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": 19253.616168904653,
-      "required_pages_per_second_upper_estimate": 0.0040100204959796215
+      "required_token_positions_per_second": 19116.71477115231,
+      "required_pages_per_second_upper_estimate": 0.003981507545160452
     },
     {
       "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": 28880.424253356978,
-      "required_pages_per_second_upper_estimate": 0.0017313373361342646
+      "required_token_positions_per_second": 28675.072156728467,
+      "required_pages_per_second_upper_estimate": 0.0017190267914958814
     },
     {
       "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": 28880.424253356978,
-      "required_pages_per_second_upper_estimate": 0.0022453259541373237
+      "required_token_positions_per_second": 28675.072156728467,
+      "required_pages_per_second_upper_estimate": 0.002229360731872872
     },
     {
       "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": 28880.424253356978,
-      "required_pages_per_second_upper_estimate": 0.3704470729384818
+      "required_token_positions_per_second": 28675.072156728467,
+      "required_pages_per_second_upper_estimate": 0.36781303673283394
     }
   ],
   "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": 0.059209600679700713,
-    "producer_p95_seconds": 0.13014462999990428,
-    "gpu_update_seconds": 0.9573266568889228,
-    "producer_batches_per_second": 16.889152916426234,
-    "gpu_batches_per_second": 1.0445755299969972,
-    "producer_headroom": 16.16843629916813,
+    "producer_batch_seconds": 0.047763477140595256,
+    "producer_p95_seconds": 0.13822026499974527,
+    "gpu_update_seconds": 0.9641824037577018,
+    "producer_batches_per_second": 20.93649918025079,
+    "gpu_batches_per_second": 1.0371481538168572,
+    "producer_headroom": 20.186604105885362,
     "sustainable": true,
-    "buffer_seconds": 60.31157938400214,
+    "buffer_seconds": 60.74349143673521,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 59,
+    "upload_pages_per_commit": 64,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

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

@@ -8,3 +8,4 @@
 {"step": 800800, "loss": 0.5392715334892273, "lr": 3e-05, "grad_norm": 0.26546578503392076, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117213898762, "tokens_per_second": 151639.58409232288, "data_wait_s": 1.4631999874836765e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 800900, "loss": 0.5483314394950867, "lr": 3e-05, "grad_norm": 0.27487786525222574, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117228365568, "tokens_per_second": 150913.1377443111, "data_wait_s": 1.7877000118460273e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 801000, "loss": 0.5693171620368958, "lr": 3e-05, "grad_norm": 0.26937283032833953, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117242932024, "tokens_per_second": 151567.21791711316, "data_wait_s": 1.225300002261065e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 801100, "loss": 0.5213466286659241, "lr": 3e-05, "grad_norm": 0.28487558318618295, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117257534817, "tokens_per_second": 152277.8665797183, "data_wait_s": 1.150700018115458e-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 sha256:522ba2ff478d0342f007cfcd8401b63f2033edf2b0e977dc1f3c5700c3b97b50
-size 21345
+oid sha256:2bd664920c1afc9ecfe93fe1d1133ee6179130e943d93983d818153cf4447cda
+size 23463