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": 76415.40992011143,
-      "required_pages_per_second_upper_estimate": 0.005286595366452623
+      "required_token_positions_per_second": 77219.31982389842,
+      "required_pages_per_second_upper_estimate": 0.005342211718924576
     },
     {
       "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": 9551.926240013929,
-      "required_pages_per_second_upper_estimate": 0.003999192049791803
+      "required_token_positions_per_second": 9652.414977987302,
+      "required_pages_per_second_upper_estimate": 0.004041264585937784
     },
     {
       "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": 19103.852480027857,
-      "required_pages_per_second_upper_estimate": 0.003978828669120656
+      "required_token_positions_per_second": 19304.829955974605,
+      "required_pages_per_second_upper_estimate": 0.004020686977227904
     },
     {
       "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": 28655.778720041784,
-      "required_pages_per_second_upper_estimate": 0.001717870179425887
+      "required_token_positions_per_second": 28957.244933961905,
+      "required_pages_per_second_upper_estimate": 0.0017359426186381425
     },
     {
       "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": 28655.778720041784,
-      "required_pages_per_second_upper_estimate": 0.0022278607520333418
+      "required_token_positions_per_second": 28957.244933961905,
+      "required_pages_per_second_upper_estimate": 0.0022512984241558932
     },
     {
       "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": 28655.778720041784,
-      "required_pages_per_second_upper_estimate": 0.3675655612426955
+      "required_token_positions_per_second": 28957.244933961905,
+      "required_pages_per_second_upper_estimate": 0.37143244614566134
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.387462980664111,
-    "producer_p95_seconds": 0.3484402269968996,
-    "gpu_update_seconds": 0.9648315709760508,
-    "producer_batches_per_second": 2.580891723606734,
-    "gpu_batches_per_second": 1.0364503298626224,
-    "producer_headroom": 2.4901258162065725,
+    "producer_batch_seconds": 0.11634113119521317,
+    "producer_p95_seconds": 0.3454606480008806,
+    "gpu_update_seconds": 0.9547869648183835,
+    "producer_batches_per_second": 8.59541238534171,
+    "gpu_batches_per_second": 1.0473540557711916,
+    "producer_headroom": 8.206787702762753,
     "sustainable": true,
-    "buffer_seconds": 60.7843889714912,
+    "buffer_seconds": 60.151578783558165,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 278712.54448252777,
+    "generation_bytes_per_second": 279524.35665745084,
     "upload_bytes_per_second": 3973949.540153322,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 59.805223949974774,
+    "estimated_full_shards_per_hour": 59.97941994469303,
     "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-v1/1791342976705245012-6808dd50/train.jsonl

@@ -354,3 +354,4 @@
 {"step": 877900, "loss": 0.5299591422080994, "lr": 3e-05, "grad_norm": 0.31637334975366893, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128500210678, "tokens_per_second": 150146.79126671597, "data_wait_s": 1.4199998986441642e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 878000, "loss": 0.5198595523834229, "lr": 3e-05, "grad_norm": 0.30138644963271394, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128514702790, "tokens_per_second": 145731.34632457333, "data_wait_s": 1.3958000636193901e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 878100, "loss": 0.5215075016021729, "lr": 3e-05, "grad_norm": 0.30821407539493095, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128529281573, "tokens_per_second": 150957.94744914758, "data_wait_s": 1.22519995784387e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 878200, "loss": 0.5371964573860168, "lr": 3e-05, "grad_norm": 0.3090578056224381, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128544042853, "tokens_per_second": 139411.23748300635, "data_wait_s": 1.1652999091893435e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:64097ff69193e7ecf4058d9bf669739a8140ea8cac0375ce8eedb228d98ac358
-size 759654
+oid sha256:8b681a58a11e61a1b83a31315b2cbe3de52f859a0a6dd8367a6c4d017744f934
+size 761772