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": 77550.79740844281,
-      "required_pages_per_second_upper_estimate": 0.00536514410735732
+      "required_token_positions_per_second": 77477.41158000537,
+      "required_pages_per_second_upper_estimate": 0.005360067105467435
     },
     {
       "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": 9693.849676055352,
-      "required_pages_per_second_upper_estimate": 0.004058612428763988
+      "required_token_positions_per_second": 9684.676447500671,
+      "required_pages_per_second_upper_estimate": 0.004054771789526939
     },
     {
       "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": 19387.699352110703,
-      "required_pages_per_second_upper_estimate": 0.004037946487030115
+      "required_token_positions_per_second": 19369.352895001342,
+      "required_pages_per_second_upper_estimate": 0.004034125403842845
     },
     {
       "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": 29081.549028166053,
-      "required_pages_per_second_upper_estimate": 0.0017433944592843192
+      "required_token_positions_per_second": 29054.029342502014,
+      "required_pages_per_second_upper_estimate": 0.001741744696148887
     },
     {
       "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": 29081.549028166053,
-      "required_pages_per_second_upper_estimate": 0.0022609625207243383
+      "required_token_positions_per_second": 29054.029342502014,
+      "required_pages_per_second_upper_estimate": 0.002258822986210264
     },
     {
       "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": 29081.549028166053,
-      "required_pages_per_second_upper_estimate": 0.37302688559877445
+      "required_token_positions_per_second": 29054.029342502014,
+      "required_pages_per_second_upper_estimate": 0.3726738926194125
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.13840916760921118,
-    "producer_p95_seconds": 0.3029120399951353,
-    "gpu_update_seconds": 0.95070589167112,
-    "producer_batches_per_second": 7.2249549453503805,
-    "gpu_batches_per_second": 1.0518500082525337,
-    "producer_headroom": 6.868807233603001,
+    "producer_batch_seconds": 0.44255099271879317,
+    "producer_p95_seconds": 0.3087062019985751,
+    "gpu_update_seconds": 0.9516063907719269,
+    "producer_batches_per_second": 2.259626611289566,
+    "gpu_batches_per_second": 1.0508546492513748,
+    "producer_headroom": 2.1502751240614635,
     "sustainable": true,
-    "buffer_seconds": 60.84517706695168,
+    "buffer_seconds": 60.90280900940332,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 336640.5475343548,
-    "upload_bytes_per_second": 5519063.1532562915,
+    "generation_bytes_per_second": 307148.11458964006,
+    "upload_bytes_per_second": 5515963.476508198,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.23522491000159,
+    "estimated_full_shards_per_hour": 65.90683534876729,
     "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": [
-    {
-      "seconds": 1.0921460530007607,
-      "bytes": 6276750,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.0008894900020096,
       "bytes": 2319506,
@@ -562,6 +556,12 @@
       "bytes": 2535870,
       "pages": 34,
       "shards": 2
+    },
+    {
+      "seconds": 1.0642618760030018,
+      "bytes": 5883611,
+      "pages": 24,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -365,3 +365,4 @@
 {"step": 521500, "loss": 0.57232266664505, "lr": 0.0003, "grad_norm": 0.16306147595004902, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76330023305, "tokens_per_second": 151847.95394713708, "data_wait_s": 1.5318000805564225e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 521600, "loss": 0.6300690174102783, "lr": 0.0003, "grad_norm": 0.20283871393591768, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76344462949, "tokens_per_second": 150974.9642086803, "data_wait_s": 1.1862000974360853e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 521700, "loss": 0.5954093933105469, "lr": 0.0003, "grad_norm": 0.20252088826755185, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76359106129, "tokens_per_second": 153328.44297819925, "data_wait_s": 1.2363998393993825e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 521800, "loss": 0.5179608464241028, "lr": 0.0003, "grad_norm": 0.15108013226515238, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76373767291, "tokens_per_second": 152419.8019010753, "data_wait_s": 8.96599522093311e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e3d8de0927e5ca7833c55397c1c0ca8da1ff703baf22e1ce24dea9fe6e67a5d2
-size 782952
+oid sha256:791e49c989e60c67d69b3863e3f79c3f130c6c0c3560b24df651c79af11bceef
+size 785070