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": 76909.8476074457,
-      "required_pages_per_second_upper_estimate": 0.005320801712915903
+      "required_token_positions_per_second": 77291.18734317599,
+      "required_pages_per_second_upper_estimate": 0.005347183680663817
     },
     {
       "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": 9613.730950930712,
-      "required_pages_per_second_upper_estimate": 0.004025068391623534
+      "required_token_positions_per_second": 9661.398417896999,
+      "required_pages_per_second_upper_estimate": 0.004045025764632416
     },
     {
       "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": 19227.461901861425,
-      "required_pages_per_second_upper_estimate": 0.004004573251888922
+      "required_token_positions_per_second": 19322.796835793997,
+      "required_pages_per_second_upper_estimate": 0.0040244290044757994
     },
     {
       "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": 28841.192852792137,
-      "required_pages_per_second_upper_estimate": 0.0017289854735732825
+      "required_token_positions_per_second": 28984.195253690992,
+      "required_pages_per_second_upper_estimate": 0.0017375582491551457
     },
     {
       "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": 28841.192852792137,
-      "required_pages_per_second_upper_estimate": 0.002242275885304102
+      "required_token_positions_per_second": 28984.195253690992,
+      "required_pages_per_second_upper_estimate": 0.002253393692972901
     },
     {
       "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": 28841.192852792137,
-      "required_pages_per_second_upper_estimate": 0.36994385465543206
+      "required_token_positions_per_second": 28984.195253690992,
+      "required_pages_per_second_upper_estimate": 0.3717781359101473
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.06153121718751464,
-    "producer_p95_seconds": 0.14132753700005196,
-    "gpu_update_seconds": 0.9586288660499483,
-    "producer_batches_per_second": 16.25191318664359,
-    "gpu_batches_per_second": 1.043156570196475,
-    "producer_headroom": 15.579553109254348,
+    "producer_batch_seconds": 0.2631649265937632,
+    "producer_p95_seconds": 0.14356579900004363,
+    "gpu_update_seconds": 0.9538991770516438,
+    "producer_batches_per_second": 3.799898462699243,
+    "gpu_batches_per_second": 1.048328821386393,
+    "producer_headroom": 3.6247200164486144,
     "sustainable": true,
-    "buffer_seconds": 60.393618561146745,
+    "buffer_seconds": 60.09564815425356,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 68,
+    "upload_pages_per_commit": 65,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 343059.2294637467,
+    "generation_bytes_per_second": 350299.43038647465,
     "upload_bytes_per_second": 4804277.543330234,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.61252463278103,
+    "estimated_full_shards_per_hour": 75.166103207547,
     "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-v1/1790989041640695387-36510d53/train.jsonl

@@ -13,3 +13,4 @@
 {"step": 558400, "loss": 0.5963554978370667, "lr": 0.0003, "grad_norm": 0.17721178056017542, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81731761513, "tokens_per_second": 152104.66692934177, "data_wait_s": 1.2724000043817796e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 558500, "loss": 0.5885961055755615, "lr": 0.0003, "grad_norm": 0.17555723889922717, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81746302396, "tokens_per_second": 118451.6538959303, "data_wait_s": 1.4738000118086347e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 558600, "loss": 0.5376635193824768, "lr": 0.0003, "grad_norm": 0.1672256278354582, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81760870349, "tokens_per_second": 151810.27200362913, "data_wait_s": 1.2439000101949205e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 558700, "loss": 0.5934765338897705, "lr": 0.0003, "grad_norm": 0.15493456042378062, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81775657961, "tokens_per_second": 153695.41540460917, "data_wait_s": 1.5023999822005862e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ccbb971fce1a8e8f975cdb45e8f2b58952998c9ed1c575ebfe4edc77383c781a
-size 31935
+oid sha256:01f21c32c1fd83803ff53739fe75567ee5032dcba7387c1afab490dcdf0fe0b8
+size 34053