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": 75250.63274878252,
-      "required_pages_per_second_upper_estimate": 0.006522534125679423
+      "required_token_positions_per_second": 75350.77851912967,
+      "required_pages_per_second_upper_estimate": 0.006531214507235979
     },
     {
       "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": 9406.329093597815,
-      "required_pages_per_second_upper_estimate": 0.003752325312588884
+      "required_token_positions_per_second": 9418.847314891209,
+      "required_pages_per_second_upper_estimate": 0.0037573190182269064
     },
     {
       "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": 18812.65818719563,
-      "required_pages_per_second_upper_estimate": 0.004351977380131792
+      "required_token_positions_per_second": 18837.694629782418,
+      "required_pages_per_second_upper_estimate": 0.004357769120497905
     },
     {
       "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": 28218.987280793444,
-      "required_pages_per_second_upper_estimate": 0.0017183882718162422
+      "required_token_positions_per_second": 28256.541944673627,
+      "required_pages_per_second_upper_estimate": 0.0017206751538124516
     },
     {
       "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": 28218.987280793444,
-      "required_pages_per_second_upper_estimate": 0.002172375756031865
+      "required_token_positions_per_second": 28256.541944673627,
+      "required_pages_per_second_upper_estimate": 0.0021752668180153247
     },
     {
       "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": 28218.987280793444,
-      "required_pages_per_second_upper_estimate": 0.35038537915235785
+      "required_token_positions_per_second": 28256.541944673627,
+      "required_pages_per_second_upper_estimate": 0.35085168239037495
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.3368259728281373,
-    "producer_p95_seconds": 0.4020481330007897,
-    "gpu_update_seconds": 0.9797658478983733,
-    "producer_batches_per_second": 2.9688921896478626,
-    "gpu_batches_per_second": 1.0206520283851797,
-    "producer_headroom": 2.908819173509196,
+    "producer_batch_seconds": 0.39334333926552745,
+    "producer_p95_seconds": 0.40755225899920333,
+    "gpu_update_seconds": 0.9784636794599582,
+    "producer_batches_per_second": 2.54230820805878,
+    "gpu_batches_per_second": 1.0220103423276052,
+    "producer_headroom": 2.4875562435784473,
     "sustainable": true,
-    "buffer_seconds": 60.74548256969914,
+    "buffer_seconds": 60.66474812651741,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 384904.2706965587,
-    "upload_bytes_per_second": 5772657.955958583,
+    "generation_bytes_per_second": 349399.9626908156,
+    "upload_bytes_per_second": 5768118.13408605,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.59149637863703,
+    "estimated_full_shards_per_hour": 74.97309837859488,
     "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": 0.9270145089976722,
-      "bytes": 5635685,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.4186761179953464,
       "bytes": 11991624,
@@ -562,6 +556,12 @@
       "bytes": 7544195,
       "pages": 32,
       "shards": 3
+    },
+    {
+      "seconds": 1.183839971999987,
+      "bytes": 6786466,
+      "pages": 32,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -342,3 +342,4 @@
 {"step": 314400, "loss": 0.572323065251112, "lr": 0.0003, "grad_norm": 0.15096272257554857, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46019294449, "tokens_per_second": 149564.04520477745, "data_wait_s": 1.708900526864454e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 314500, "loss": 0.5538896843791008, "lr": 0.0003, "grad_norm": 0.14577745322222035, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46033954052, "tokens_per_second": 149969.83623488643, "data_wait_s": 9.750001481734216e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 314600, "loss": 0.583943635225296, "lr": 0.0003, "grad_norm": 0.13831777217287192, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46048707348, "tokens_per_second": 150116.10642143327, "data_wait_s": 1.1046999134123325e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 314700, "loss": 0.6050030756741762, "lr": 0.0003, "grad_norm": 0.16552769369539239, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46063448200, "tokens_per_second": 150341.50362639906, "data_wait_s": 1.4864002878312021e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c909374206da45b5f909a4279eb0ddd433f69eead0d5dbd05f73de3d250c015f
-size 733455
+oid sha256:e8853e661ef431728314525ab4d72195907d85a0ba5339bee382923b111773c1
+size 735573