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": 76236.02722997342,
-      "required_pages_per_second_upper_estimate": 0.00527418525572369
+      "required_token_positions_per_second": 77211.33312348135,
+      "required_pages_per_second_upper_estimate": 0.005341659180458026
     },
     {
       "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": 9529.503403746678,
-      "required_pages_per_second_upper_estimate": 0.0039898040764887715
+      "required_token_positions_per_second": 9651.416640435169,
+      "required_pages_per_second_upper_estimate": 0.004040846602852364
     },
     {
       "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": 19059.006807493355,
-      "required_pages_per_second_upper_estimate": 0.003969488498191634
+      "required_token_positions_per_second": 19302.833280870338,
+      "required_pages_per_second_upper_estimate": 0.004020271122459548
     },
     {
       "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": 28588.51021124003,
-      "required_pages_per_second_upper_estimate": 0.0017138375350362874
+      "required_token_positions_per_second": 28954.249921305505,
+      "required_pages_per_second_upper_estimate": 0.0017357630722025109
     },
     {
       "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": 28588.51021124003,
-      "required_pages_per_second_upper_estimate": 0.002222630921356906
+      "required_token_positions_per_second": 28954.249921305505,
+      "required_pages_per_second_upper_estimate": 0.0022510655750955263
     },
     {
       "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": 28588.51021124003,
-      "required_pages_per_second_upper_estimate": 0.366702713039084
+      "required_token_positions_per_second": 28954.249921305505,
+      "required_pages_per_second_upper_estimate": 0.3713940293390991
     }
   ],
   "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": 0.08183997820320599,
-    "producer_p95_seconds": 0.2172023929997522,
-    "gpu_update_seconds": 0.9671018110321029,
-    "producer_batches_per_second": 12.218967086196342,
-    "gpu_batches_per_second": 1.0340172964134848,
-    "producer_headroom": 11.81698519800214,
+    "producer_batch_seconds": 0.20718520321879907,
+    "producer_p95_seconds": 0.22391430599964224,
+    "gpu_update_seconds": 0.9548857274888574,
+    "producer_batches_per_second": 4.826599508382578,
+    "gpu_batches_per_second": 1.0472457292138855,
+    "producer_headroom": 4.608850982859258,
     "sustainable": true,
-    "buffer_seconds": 60.927414095022485,
+    "buffer_seconds": 60.15780083179801,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 375956.20468514727,
-    "upload_bytes_per_second": 4248779.331870772,
+    "generation_bytes_per_second": 357129.03535423434,
+    "upload_bytes_per_second": 4280977.133276105,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.6714497129041,
+    "estimated_full_shards_per_hour": 76.6315774485614,
     "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": 1.00537013399844,
-      "bytes": 7625189,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.1632388540001557,
       "bytes": 4899738,
@@ -562,6 +556,12 @@
       "bytes": 7289548,
       "pages": 26,
       "shards": 3
+    },
+    {
+      "seconds": 1.2270069759997568,
+      "bytes": 11800938,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -160,3 +160,4 @@
 {"step": 701000, "loss": 0.5181252956390381, "lr": 8e-05, "grad_norm": 0.2046413997117008, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102607411373, "tokens_per_second": 153609.74864793487, "data_wait_s": 1.1185000403202139e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 701100, "loss": 0.5583337545394897, "lr": 8e-05, "grad_norm": 0.20807786757317906, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102621901511, "tokens_per_second": 134787.57889448781, "data_wait_s": 1.7554000805830583e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 701200, "loss": 0.5611851215362549, "lr": 8e-05, "grad_norm": 0.21945404642932054, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102636378349, "tokens_per_second": 150698.8858345956, "data_wait_s": 1.5719999282737263e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 701300, "loss": 0.5354167222976685, "lr": 8e-05, "grad_norm": 0.21774947897899255, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102650876683, "tokens_per_second": 151278.39324420216, "data_wait_s": 1.918900125019718e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4e1e308d440fcf68a1be32a5646794f0d10c35d1f3ff238e71d7a59d93c40d9b
-size 345630
+oid sha256:aadcedb8ce8455217369bfacab7fd9f9b37d5ea8a833bc1f9c31a49752fa0cc5
+size 347748