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": 75215.28715122174,
-      "required_pages_per_second_upper_estimate": 0.006519470458865453
+      "required_token_positions_per_second": 75754.97546785271,
+      "required_pages_per_second_upper_estimate": 0.006566249274323492
     },
     {
       "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": 9401.910893902717,
-      "required_pages_per_second_upper_estimate": 0.0037505628266725375
+      "required_token_positions_per_second": 9469.371933481589,
+      "required_pages_per_second_upper_estimate": 0.0037774740439929745
     },
     {
       "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": 18803.821787805435,
-      "required_pages_per_second_upper_estimate": 0.004349933234648188
+      "required_token_positions_per_second": 18938.743866963177,
+      "required_pages_per_second_upper_estimate": 0.00438114508311384
     },
     {
       "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": 28205.732681708152,
-      "required_pages_per_second_upper_estimate": 0.001717581136273455
+      "required_token_positions_per_second": 28408.115800444764,
+      "required_pages_per_second_upper_estimate": 0.0017299052063823525
     },
     {
       "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": 28205.732681708152,
-      "required_pages_per_second_upper_estimate": 0.0021713553803032006
+      "required_token_positions_per_second": 28408.115800444764,
+      "required_pages_per_second_upper_estimate": 0.0021869353930158743
     },
     {
       "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": 28205.732681708152,
-      "required_pages_per_second_upper_estimate": 0.35022080139200806
+      "required_token_positions_per_second": 28408.115800444764,
+      "required_pages_per_second_upper_estimate": 0.3527337223939899
     }
   ],
   "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": 0.07965512064058089,
-    "producer_p95_seconds": 0.28817886099932366,
-    "gpu_update_seconds": 0.9802262650645537,
-    "producer_batches_per_second": 12.554120713873385,
-    "gpu_batches_per_second": 1.0201726230363193,
-    "producer_headroom": 12.305878858529656,
+    "producer_batch_seconds": 0.14952971255462444,
+    "producer_p95_seconds": 0.31884578300014255,
+    "gpu_update_seconds": 0.9732430054219624,
+    "producer_batches_per_second": 6.6876340689459415,
+    "gpu_batches_per_second": 1.0274926143100682,
+    "producer_headroom": 6.508693080423255,
     "sustainable": true,
-    "buffer_seconds": 60.774028434002325,
+    "buffer_seconds": 60.34106633616167,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 357382.8243680251,
-    "upload_bytes_per_second": 5273725.942908658,
+    "generation_bytes_per_second": 361878.9702530805,
+    "upload_bytes_per_second": 5243018.040967091,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.68603466301504,
+    "estimated_full_shards_per_hour": 77.65080290502844,
     "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": 0.9170985109994945,
-      "bytes": 6599086,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.249970442000631,
       "bytes": 7464066,
@@ -562,6 +556,12 @@
       "bytes": 6080871,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.1276710530000855,
+      "bytes": 5237494,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -245,3 +245,4 @@
 {"step": 339500, "loss": 0.5592187382280827, "lr": 0.0003, "grad_norm": 0.13438206562887653, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49693527275, "tokens_per_second": 147942.16319160472, "data_wait_s": 1.5681998775107786e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 339600, "loss": 0.5828502215445042, "lr": 0.0003, "grad_norm": 0.14440584274616805, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49708161255, "tokens_per_second": 148282.36220059928, "data_wait_s": 1.401100234943442e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 339700, "loss": 0.5540956780314445, "lr": 0.0003, "grad_norm": 0.13777634286151783, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49722893086, "tokens_per_second": 150541.83519850168, "data_wait_s": 1.3155997294234112e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 339800, "loss": 0.6158789154142141, "lr": 0.0003, "grad_norm": 0.13946108512276084, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49737582528, "tokens_per_second": 149685.63441450475, "data_wait_s": 1.3250999472802505e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3c8095752f439fd4cdd895aaafacf066719d77695f169df058085feb4e058d5b
-size 526443
+oid sha256:a0f6c410e8649b1986ad8e079dd4d845617b78e11aed55dd70c202c94cf1db0a
+size 528561