Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75374.46762121329,
-      "required_pages_per_second_upper_estimate": 0.006533267818565108
+      "required_token_positions_per_second": 75202.34237512595,
+      "required_pages_per_second_upper_estimate": 0.006518348438481705
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9421.808452651661,
-      "required_pages_per_second_upper_estimate": 0.0037585002603525057
+      "required_token_positions_per_second": 9400.292796890744,
+      "required_pages_per_second_upper_estimate": 0.0037499173435817552
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18843.616905303323,
-      "required_pages_per_second_upper_estimate": 0.004359139134711178
+      "required_token_positions_per_second": 18800.58559378149,
+      "required_pages_per_second_upper_estimate": 0.004349184598115818
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28265.425357954984,
-      "required_pages_per_second_upper_estimate": 0.0017212161070736189
+      "required_token_positions_per_second": 28200.87839067223,
+      "required_pages_per_second_upper_estimate": 0.0017172855354178642
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28265.425357954984,
-      "required_pages_per_second_upper_estimate": 0.0021759506877605853
+      "required_token_positions_per_second": 28200.87839067223,
+      "required_pages_per_second_upper_estimate": 0.002170981683541718
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28265.425357954984,
-      "required_pages_per_second_upper_estimate": 0.3509619846524577
+      "required_token_positions_per_second": 28200.87839067223,
+      "required_pages_per_second_upper_estimate": 0.3501605273436089
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.08587130319531866,
-    "producer_p95_seconds": 0.22176854100143828,
-    "gpu_update_seconds": 0.9781561625153037,
-    "producer_batches_per_second": 11.645333921688005,
-    "gpu_batches_per_second": 1.022331646338071,
-    "producer_headroom": 11.390955140047632,
+    "producer_batch_seconds": 0.14161902912501034,
+    "producer_p95_seconds": 0.22650092399999266,
+    "gpu_update_seconds": 0.9803949939780917,
+    "producer_batches_per_second": 7.061197963144326,
+    "gpu_batches_per_second": 1.0199970482737353,
+    "producer_headroom": 6.9227631345549945,
     "sustainable": true,
-    "buffer_seconds": 60.64568207594883,
+    "buffer_seconds": 60.78448962664169,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 380686.652612275,
-    "upload_bytes_per_second": 5269207.496906731,
+    "generation_bytes_per_second": 372914.80463842815,
+    "upload_bytes_per_second": 5250636.950515227,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.68649371887385,
+    "estimated_full_shards_per_hour": 80.01883606304773,
     "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": 1.1525904230002197,
-      "bytes": 7408784,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.0340140519992929,
       "bytes": 5739884,
@@ -562,6 +556,12 @@
       "bytes": 4975022,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.1054393590002292,
+      "bytes": 5686965,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -141,3 +141,4 @@
 {"step": 209100, "loss": 0.6039793342351913, "lr": 0.0003, "grad_norm": 0.1505738887339764, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30604528995, "tokens_per_second": 146527.98548877737, "data_wait_s": 2.1677000404451974e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 209200, "loss": 0.5919628143310547, "lr": 0.0003, "grad_norm": 0.15929568507064515, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30619022799, "tokens_per_second": 146032.3033093075, "data_wait_s": 1.7292000848101452e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 209300, "loss": 0.61921376734972, "lr": 0.0003, "grad_norm": 0.14681122395405305, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30633559690, "tokens_per_second": 147517.66940943303, "data_wait_s": 1.7534999642521143e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 209400, "loss": 0.6080213040113449, "lr": 0.0003, "grad_norm": 0.14197261232353933, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 30648187358, "tokens_per_second": 147741.7675434638, "data_wait_s": 1.1895001080119982e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:affd71d52655e4638e84780b2279801cdd8c2a21a4d8c0e4711f36200dcdd47e
-size 304605
+oid sha256:06c1a3fd925c0df5ef50f50f172c01cd93c39a54a76030420878971232455d31
+size 306723