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": 75087.34730056021,
-      "required_pages_per_second_upper_estimate": 0.006508380956870698
+      "required_token_positions_per_second": 75181.52378642866,
+      "required_pages_per_second_upper_estimate": 0.006516543935977654
     },
     {
       "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": 9385.918412570027,
-      "required_pages_per_second_upper_estimate": 0.0037441831867600236
+      "required_token_positions_per_second": 9397.690473303583,
+      "required_pages_per_second_upper_estimate": 0.0037488792377946317
     },
     {
       "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": 18771.836825140053,
-      "required_pages_per_second_upper_estimate": 0.004342534076451669
+      "required_token_positions_per_second": 18795.380946607165,
+      "required_pages_per_second_upper_estimate": 0.0043479805945907
     },
     {
       "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": 28157.755237710084,
-      "required_pages_per_second_upper_estimate": 0.0017146595616521674
+      "required_token_positions_per_second": 28193.071419910746,
+      "required_pages_per_second_upper_estimate": 0.0017168101318584996
     },
     {
       "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": 28157.755237710084,
-      "required_pages_per_second_upper_estimate": 0.002167661943854164
+      "required_token_positions_per_second": 28193.071419910746,
+      "required_pages_per_second_upper_estimate": 0.0021703806813214203
     },
     {
       "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": 28157.755237710084,
-      "required_pages_per_second_upper_estimate": 0.3496250821077279
+      "required_token_positions_per_second": 28193.071419910746,
+      "required_pages_per_second_upper_estimate": 0.3500635908950016
     }
   ],
   "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": 0.10815951453912476,
-    "producer_p95_seconds": 0.30435760999898775,
-    "gpu_update_seconds": 0.9818964532716942,
-    "producer_batches_per_second": 9.245603627762845,
-    "gpu_batches_per_second": 1.0184373277528243,
-    "producer_headroom": 9.078225410456247,
+    "producer_batch_seconds": 0.06835356326570263,
+    "producer_p95_seconds": 0.22755495999808772,
+    "gpu_update_seconds": 0.9806664761071119,
+    "producer_batches_per_second": 14.629815216989048,
+    "gpu_batches_per_second": 1.0197146780928366,
+    "producer_headroom": 14.346969334942852,
     "sustainable": true,
-    "buffer_seconds": 60.87758010284504,
+    "buffer_seconds": 60.80132151864094,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 381845.801104078,
-    "upload_bytes_per_second": 5971096.05078665,
+    "generation_bytes_per_second": 382498.37677806267,
+    "upload_bytes_per_second": 6025430.015434966,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.93522000161892,
+    "estimated_full_shards_per_hour": 82.07524755007181,
     "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": 1.07777969900053,
-      "bytes": 7140650,
-      "pages": 25,
-      "shards": 3
-    },
     {
       "seconds": 1.360217057999762,
       "bytes": 16210349,
@@ -562,6 +556,12 @@
       "bytes": 906815,
       "pages": 27,
       "shards": 1
+    },
+    {
+      "seconds": 1.218643701999099,
+      "bytes": 11866314,
+      "pages": 38,
+      "shards": 5
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -142,3 +142,4 @@
 {"step": 99200, "loss": 0.6560120675712824, "lr": 0.0003, "grad_norm": 0.1448230448036412, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14515937038, "tokens_per_second": 149260.46656382762, "data_wait_s": 1.048400008585304e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 99300, "loss": 0.5981705710291862, "lr": 0.0003, "grad_norm": 0.14313228307489784, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14530706160, "tokens_per_second": 149255.5627441101, "data_wait_s": 1.6499998309882358e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 99400, "loss": 0.6243044082075357, "lr": 0.0003, "grad_norm": 0.13784995364339628, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14545501583, "tokens_per_second": 150002.45710737514, "data_wait_s": 9.878000128082931e-06, "gpu_peak_gb": 85.46991729736328}
+{"step": 99500, "loss": 0.6166529804468155, "lr": 0.0003, "grad_norm": 0.13084167545458994, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14560117029, "tokens_per_second": 148304.48641367702, "data_wait_s": 1.5222998626995832e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4ea4305fce7b3746d889615a8135fe678b4285d6b230b710990db313b40001ac
-size 306723
+oid sha256:6f6da1fdf097bb04936a0f6a6f1c93aef113ca6473383af898f59e29ccfdc5f4
+size 308841