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": 74662.79907872532,
-      "required_pages_per_second_upper_estimate": 0.006471582193009681
+      "required_token_positions_per_second": 73928.83451130315,
+      "required_pages_per_second_upper_estimate": 0.006407964004521717
     },
     {
       "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": 9332.849884840665,
-      "required_pages_per_second_upper_estimate": 0.003723013357603584
+      "required_token_positions_per_second": 9241.104313912894,
+      "required_pages_per_second_upper_estimate": 0.0036864146776419715
     },
     {
       "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": 18665.69976968133,
-      "required_pages_per_second_upper_estimate": 0.0043179811366196716
+      "required_token_positions_per_second": 18482.20862782579,
+      "required_pages_per_second_upper_estimate": 0.004275533690285193
     },
     {
       "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": 27998.549654521994,
-      "required_pages_per_second_upper_estimate": 0.0017049647769231534
+      "required_token_positions_per_second": 27723.31294173868,
+      "required_pages_per_second_upper_estimate": 0.001688204305170079
     },
     {
       "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": 27998.549654521994,
-      "required_pages_per_second_upper_estimate": 0.0021554058573510842
+      "required_token_positions_per_second": 27723.31294173868,
+      "required_pages_per_second_upper_estimate": 0.0021342173732970394
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27998.549654521994,
-      "required_pages_per_second_upper_estimate": 0.34764828159134303
+      "required_token_positions_per_second": 27723.31294173868,
+      "required_pages_per_second_upper_estimate": 0.3442307627765956
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.1735889495543006,
-    "producer_p95_seconds": 0.4106779829962761,
-    "gpu_update_seconds": 0.9874797209552825,
-    "producer_batches_per_second": 5.760735361136502,
-    "gpu_batches_per_second": 1.0126790239627457,
-    "producer_headroom": 5.688609346912302,
+    "producer_batch_seconds": 0.5142520451092878,
+    "producer_p95_seconds": 0.40062544199463446,
+    "gpu_update_seconds": 0.9972834075820248,
+    "producer_batches_per_second": 1.944571751362665,
+    "gpu_batches_per_second": 1.0027239923950622,
+    "producer_headroom": 1.9392891424867045,
     "sustainable": true,
-    "buffer_seconds": 60.23626297827223,
+    "buffer_seconds": 60.834287862503515,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 263058.4646103507,
+    "generation_bytes_per_second": 243763.52235258414,
     "upload_bytes_per_second": 6032254.231295674,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 56.446222817734636,
+    "estimated_full_shards_per_hour": 52.30597737248557,
     "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-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -367,3 +367,4 @@
 {"step": 151700, "loss": 0.6355745941400528, "lr": 0.0003, "grad_norm": 0.1451971084639051, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22201131978, "tokens_per_second": 149029.01259873327, "data_wait_s": 9.254996257368475e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 151800, "loss": 0.6344895455986261, "lr": 0.0003, "grad_norm": 0.16112494940064362, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22215780946, "tokens_per_second": 149172.94995248725, "data_wait_s": 1.5428995538968593e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 151900, "loss": 0.6639365497976542, "lr": 0.0003, "grad_norm": 0.14157858284681954, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22230487947, "tokens_per_second": 128559.01434671022, "data_wait_s": 1.8451006326358765e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 152000, "loss": 0.6232686080038548, "lr": 0.0003, "grad_norm": 0.1419686044015527, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22245102282, "tokens_per_second": 149083.7484558147, "data_wait_s": 9.306000720243901e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b6f24753252e58d7a95a5b84edf6c7ac3d25cdc549551a992744710feba38566
-size 787188
+oid sha256:fdc6d6974a38562e9e6e14c3bf8920b78ec4701b3724f7985df8ec3b773f6e85
+size 789306