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": 75942.36716961404,
-      "required_pages_per_second_upper_estimate": 0.006582491912092238
+      "required_token_positions_per_second": 75827.8216053968,
+      "required_pages_per_second_upper_estimate": 0.006572563392898967
     },
     {
       "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": 9492.795896201755,
-      "required_pages_per_second_upper_estimate": 0.003786818212941501
+      "required_token_positions_per_second": 9478.4777006746,
+      "required_pages_per_second_upper_estimate": 0.003781106470669619
     },
     {
       "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": 18985.59179240351,
-      "required_pages_per_second_upper_estimate": 0.004391982526169133
+      "required_token_positions_per_second": 18956.9554013492,
+      "required_pages_per_second_upper_estimate": 0.0043853580010819885
     },
     {
       "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": 28478.387688605264,
-      "required_pages_per_second_upper_estimate": 0.0017341843956832235
+      "required_token_positions_per_second": 28435.433102023806,
+      "required_pages_per_second_upper_estimate": 0.0017315686867257114
     },
     {
       "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": 28478.387688605264,
-      "required_pages_per_second_upper_estimate": 0.002192345117491506
+      "required_token_positions_per_second": 28435.433102023806,
+      "required_pages_per_second_upper_estimate": 0.0021890383545105593
     },
     {
       "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": 28478.387688605264,
-      "required_pages_per_second_upper_estimate": 0.3536062640600626
+      "required_token_positions_per_second": 28435.433102023806,
+      "required_pages_per_second_upper_estimate": 0.3530729118544744
     }
   ],
   "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": 0.1809595380705673,
-    "producer_p95_seconds": 0.3552755540004,
-    "gpu_update_seconds": 0.9708414781874215,
-    "producer_batches_per_second": 5.52609721853975,
-    "gpu_batches_per_second": 1.030034276931614,
-    "producer_headroom": 5.36496439225453,
+    "producer_batch_seconds": 0.1338761316245609,
+    "producer_p95_seconds": 0.3459697930011316,
+    "gpu_update_seconds": 0.972308032052877,
+    "producer_batches_per_second": 7.469591389183373,
+    "gpu_batches_per_second": 1.0284806532850046,
+    "producer_headroom": 7.2627437038560005,
     "sustainable": true,
-    "buffer_seconds": 60.192171647620135,
+    "buffer_seconds": 60.283097987278374,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 498561.98757794074,
-    "upload_bytes_per_second": 4039224.985918398,
+    "generation_bytes_per_second": 498837.9622441903,
+    "upload_bytes_per_second": 4031531.5687924474,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 106.97979660514514,
+    "estimated_full_shards_per_hour": 107.03901434416085,
     "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": 0.9711113170014869,
-      "bytes": 6125935,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.0254750570020406,
       "bytes": 8509608,
@@ -562,6 +556,12 @@
       "bytes": 12066636,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 0.828995023999596,
+      "bytes": 4743589,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -328,3 +328,4 @@
 {"step": 72800, "loss": 0.6211061533540487, "lr": 0.0003, "grad_norm": 0.15273745097921554, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10654408420, "tokens_per_second": 150021.76071150662, "data_wait_s": 9.208997653331608e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 72900, "loss": 0.6699622832238674, "lr": 0.0003, "grad_norm": 0.1450948861588816, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10668920705, "tokens_per_second": 148963.6484958983, "data_wait_s": 1.0960997315123677e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 73000, "loss": 0.6199126038700342, "lr": 0.0003, "grad_norm": 0.12562120347064035, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10683593832, "tokens_per_second": 149884.9240272704, "data_wait_s": 8.317998435813934e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 73100, "loss": 0.6249210871756077, "lr": 0.0003, "grad_norm": 0.13517386636139803, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10698270266, "tokens_per_second": 149889.29134254056, "data_wait_s": 8.770999556872994e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d7139dffcb3d12f714f256c422b8161099083274423550e3ac8336275b03a041
-size 703803
+oid sha256:b20958950ac22045aebacb9d398ae34a78c457905500847f5d8dcead51898e8b
+size 705921