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": 74989.30746794921,
-      "required_pages_per_second_upper_estimate": 0.006499883112659661
+      "required_token_positions_per_second": 74278.66020354918,
+      "required_pages_per_second_upper_estimate": 0.006438286008900491
     },
     {
       "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": 9373.663433493652,
-      "required_pages_per_second_upper_estimate": 0.003739294492378192
+      "required_token_positions_per_second": 9284.832525443648,
+      "required_pages_per_second_upper_estimate": 0.003703858515016614
     },
     {
       "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": 18747.326866987303,
-      "required_pages_per_second_upper_estimate": 0.004336864128993562
+      "required_token_positions_per_second": 18569.665050887295,
+      "required_pages_per_second_upper_estimate": 0.004295765194579996
     },
     {
       "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": 28120.990300480953,
-      "required_pages_per_second_upper_estimate": 0.0017124207698656316
+      "required_token_positions_per_second": 27854.49757633094,
+      "required_pages_per_second_upper_estimate": 0.0016961927611441625
     },
     {
       "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": 28120.990300480953,
-      "required_pages_per_second_upper_estimate": 0.0021648316772144055
+      "required_token_positions_per_second": 27854.49757633094,
+      "required_pages_per_second_upper_estimate": 0.0021443163296102625
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28120.990300480953,
-      "required_pages_per_second_upper_estimate": 0.34916858463167183
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27854.49757633094,
+      "required_pages_per_second_upper_estimate": 0.34585963689150256
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.19136436798433465,
-    "producer_p95_seconds": 0.24719173000084993,
-    "gpu_update_seconds": 0.983180169139603,
-    "producer_batches_per_second": 5.225633228030525,
-    "gpu_batches_per_second": 1.017107577419016,
-    "producer_headroom": 5.137738960996582,
+    "producer_batch_seconds": 0.08179642717985303,
+    "producer_p95_seconds": 0.2388649650001753,
+    "gpu_update_seconds": 0.9925865625195691,
+    "producer_batches_per_second": 12.225472853492876,
+    "gpu_batches_per_second": 1.0074688070142848,
+    "producer_headroom": 12.134840074824801,
     "sustainable": true,
-    "buffer_seconds": 60.957170486655386,
+    "buffer_seconds": 60.547780313693714,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 420252.38688639435,
-    "upload_bytes_per_second": 4075515.1414463604,
+    "generation_bytes_per_second": 419919.2843856162,
+    "upload_bytes_per_second": 4054731.184984265,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.17637925094483,
+    "estimated_full_shards_per_hour": 90.1049032085072,
     "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": 1.2710061699999642,
-      "bytes": 13577169,
-      "pages": 42,
-      "shards": 4
-    },
     {
       "seconds": 0.8450371549988631,
       "bytes": 895470,
@@ -562,6 +556,12 @@
       "bytes": 13478878,
       "pages": 47,
       "shards": 4
+    },
+    {
+      "seconds": 1.1504248440014635,
+      "bytes": 10949848,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -153,3 +153,4 @@
 {"step": 330300, "loss": 0.6037442348897457, "lr": 0.0003, "grad_norm": 0.17942496505797634, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48346976699, "tokens_per_second": 140060.01440860407, "data_wait_s": 1.7827998817665502e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 330400, "loss": 0.6018347283825278, "lr": 0.0003, "grad_norm": 0.1707738632691945, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48361619115, "tokens_per_second": 148133.64923696552, "data_wait_s": 1.0896999810938723e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 330500, "loss": 0.5742582809180021, "lr": 0.0003, "grad_norm": 0.19433099567572845, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48376117798, "tokens_per_second": 147441.66743168022, "data_wait_s": 1.3945998944109306e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 330600, "loss": 0.6046997979283333, "lr": 0.0003, "grad_norm": 0.19152793513634442, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48390796240, "tokens_per_second": 149141.94140897636, "data_wait_s": 1.0780999218695797e-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:dda963853b83303a8c14af4741e68e493f751319387fe58c18f45d0a9abe5aad
-size 330804
+oid sha256:b9a738fbab92411695f119f5de4e1c717a98c36e2abd82f6b8dd1335220f67d1
+size 332922