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": 76782.95460242347,
-      "required_pages_per_second_upper_estimate": 0.005312022960396129
+      "required_token_positions_per_second": 76429.93213561276,
+      "required_pages_per_second_upper_estimate": 0.005287600047017188
     },
     {
       "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": 9597.869325302934,
-      "required_pages_per_second_upper_estimate": 0.004018427460201592
+      "required_token_positions_per_second": 9553.741516951595,
+      "required_pages_per_second_upper_estimate": 0.003999952068338311
     },
     {
       "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": 19195.73865060587,
-      "required_pages_per_second_upper_estimate": 0.003997966135250784
+      "required_token_positions_per_second": 19107.48303390319,
+      "required_pages_per_second_upper_estimate": 0.003979584817748743
     },
     {
       "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": 28793.607975908802,
-      "required_pages_per_second_upper_estimate": 0.0017261328328620265
+      "required_token_positions_per_second": 28661.224550854786,
+      "required_pages_per_second_upper_estimate": 0.0017181966486678243
     },
     {
       "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": 28793.607975908802,
-      "required_pages_per_second_upper_estimate": 0.002238576370430176
+      "required_token_positions_per_second": 28661.224550854786,
+      "required_pages_per_second_upper_estimate": 0.002228284141425374
     },
     {
       "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": 28793.607975908802,
-      "required_pages_per_second_upper_estimate": 0.36933348694743273
+      "required_token_positions_per_second": 28661.224550854786,
+      "required_pages_per_second_upper_estimate": 0.3676354145130871
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.1344763617499325,
-    "producer_p95_seconds": 0.2979646759995376,
-    "gpu_update_seconds": 0.960213114769524,
-    "producer_batches_per_second": 7.436251152150924,
-    "gpu_batches_per_second": 1.041435473665683,
-    "producer_headroom": 7.1403858810153,
+    "producer_batch_seconds": 0.08286259875751512,
+    "producer_p95_seconds": 0.29357527299725916,
+    "gpu_update_seconds": 0.9646482463072371,
+    "producer_batches_per_second": 12.068170863508023,
+    "gpu_batches_per_second": 1.0366473000164491,
+    "producer_headroom": 11.64153985961911,
     "sustainable": true,
-    "buffer_seconds": 60.49342623048001,
+    "buffer_seconds": 60.77283951735594,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 394747.92881338083,
-    "upload_bytes_per_second": 5036255.955241772,
+    "generation_bytes_per_second": 394663.3219514051,
+    "upload_bytes_per_second": 5032097.611629188,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.7037162618739,
+    "estimated_full_shards_per_hour": 84.68556159884085,
     "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": 1.364633949000563,
-      "bytes": 7290084,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 1.0805251250021684,
       "bytes": 6054722,
@@ -562,6 +556,12 @@
       "bytes": 5130348,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.503245963001973,
+      "bytes": 7636802,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -355,3 +355,4 @@
 {"step": 755500, "loss": 0.5137624144554138, "lr": 5e-05, "grad_norm": 0.2398494563840914, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110584181302, "tokens_per_second": 152391.71439916862, "data_wait_s": 1.075599720934406e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 755600, "loss": 0.5737913250923157, "lr": 5e-05, "grad_norm": 0.2448262742735738, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110598971512, "tokens_per_second": 152754.81268828092, "data_wait_s": 1.0906995157711208e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 755700, "loss": 0.5509876012802124, "lr": 5e-05, "grad_norm": 0.24811867638996113, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110613679862, "tokens_per_second": 151889.6556943709, "data_wait_s": 1.3719996786676347e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 755800, "loss": 0.5624434351921082, "lr": 5e-05, "grad_norm": 0.25050675376137443, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110628188586, "tokens_per_second": 150407.52231608026, "data_wait_s": 9.04199987417087e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7c9387d7d725d8bd0fe0ba7707761b91d0babaac06f5c4a89270b0d88b62f998
-size 761772
+oid sha256:3d64b4417eb6089bc02b5287cf38a8f5eb91c3c838f41b7b4f6cfdd7bd5aa2c4
+size 763890